On Sun, May 22, 2022 at 07:35:59PM +0800, Muchun Song wrote:
> On Mon, May 09, 2022 at 11:38:20AM -0700, Roman Gushchin wrote:
> > Add a scan interface which allows to trigger scanning of a particular
> > shrinker and specify memcg and numa node. It's useful for testing,
> > debugging and profiling of a specific scan_objects() callback.
> > Unlike alternatives (creating a real memory pressure and dropping
> > caches via /proc/sys/vm/drop_caches) this interface allows to interact
> > with only one shrinker at once. Also, if a shrinker is misreporting
> > the number of objects (as some do), it doesn't affect scanning.
> >
> > Signed-off-by: Roman Gushchin <roman.gushchin@xxxxxxxxx>
> > ---
> > .../admin-guide/mm/shrinker_debugfs.rst | 39 +++++++++-
> > mm/shrinker_debug.c | 73 +++++++++++++++++++
> > 2 files changed, 108 insertions(+), 4 deletions(-)
> >
> > diff --git a/Documentation/admin-guide/mm/shrinker_debugfs.rst b/Documentation/admin-guide/mm/shrinker_debugfs.rst
> > index 6783f8190e63..8fecf81d60ee 100644
> > --- a/Documentation/admin-guide/mm/shrinker_debugfs.rst
> > +++ b/Documentation/admin-guide/mm/shrinker_debugfs.rst
> > @@ -5,14 +5,16 @@ Shrinker Debugfs Interface
> > ==========================
> >
> > Shrinker debugfs interface provides a visibility into the kernel memory
> > -shrinkers subsystem and allows to get information about individual shrinkers.
> > +shrinkers subsystem and allows to get information about individual shrinkers
> > +and interact with them.
> >
> > For each shrinker registered in the system a directory in **<debugfs>/shrinker/**
> > is created. The directory's name is composed from the shrinker's name and an
> > unique id: e.g. *kfree_rcu-0* or *sb-xfs:vda1-36*.
> >
> > -Each shrinker directory contains the **count** file, which allows to trigger
> > -the *count_objects()* callback for each memcg and numa node (if applicable).
> > +Each shrinker directory contains **count** and **scan** files, which allow to
> > +trigger *count_objects()* and *scan_objects()* callbacks for each memcg and
> > +numa node (if applicable).
> >
> > Usage:
> > ------
> > @@ -43,7 +45,7 @@ Usage:
> >
> > $ cd sb-btrfs\:vda2-24/
> > $ ls
> > - count
> > + count scan
> >
> > 3. *Count objects*
> >
> > @@ -98,3 +100,32 @@ Usage:
> > 2877 84 0
> > 293 1 0
> > 735 8 0
> > +
> > +4. *Scan objects*
> > +
> > + The expected input format::
> > +
> > + <cgroup inode id> <numa id> <number of objects to scan>
> > +
> > + For a non-memcg-aware shrinker or on a system with no memory
> > + cgrups **0** should be passed as cgroup id.
> > + ::
> > +
> > + $ cd /sys/kernel/debug/shrinker/
> > + $ cd sb-btrfs\:vda2-24/
> > +
> > + $ cat count | head -n 5
> > + 1 212 0
> > + 21 97 0
> > + 55 802 5
> > + 2367 2 0
> > + 225 13 0
> > +
> > + $ echo "55 0 200" > scan
> > +
> > + $ cat count | head -n 5
> > + 1 212 0
> > + 21 96 0
> > + 55 752 5
> > + 2367 2 0
> > + 225 13 0
> > diff --git a/mm/shrinker_debug.c b/mm/shrinker_debug.c
> > index 28b1c1ab60ef..8f67fef5a643 100644
> > --- a/mm/shrinker_debug.c
> > +++ b/mm/shrinker_debug.c
> > @@ -101,6 +101,77 @@ static int shrinker_debugfs_count_show(struct seq_file *m, void *v)
> > }
> > DEFINE_SHOW_ATTRIBUTE(shrinker_debugfs_count);
> >
> > +static int shrinker_debugfs_scan_open(struct inode *inode, struct file *file)
> > +{
> > + file->private_data = inode->i_private;
> > + return nonseekable_open(inode, file);
> > +}
> > +
> > +static ssize_t shrinker_debugfs_scan_write(struct file *file,
> > + const char __user *buf,
> > + size_t size, loff_t *pos)
> > +{
> > + struct shrinker *shrinker = (struct shrinker *)file->private_data;
>
> Seems we could drop the cast since ->private_data is void * type.
Yep, fixed. Thanks!
>
> > + unsigned long nr_to_scan = 0, ino;
> > + struct shrink_control sc = {
> > + .gfp_mask = GFP_KERNEL,
> > + };
> > + struct mem_cgroup *memcg = NULL;
> > + int nid;
> > + char kbuf[72];
> > + int read_len = size < (sizeof(kbuf) - 1) ? size : (sizeof(kbuf) - 1);
> > + ssize_t ret;
> > +
> > + if (copy_from_user(kbuf, buf, read_len))
> > + return -EFAULT;
> > + kbuf[read_len] = '\0';
> > +
> > + if (sscanf(kbuf, "%lu %d %lu", &ino, &nid, &nr_to_scan) < 2)
> > + return -EINVAL;
> > +
> > + if (nid < 0 || nid >= nr_node_ids)
> > + return -EINVAL;
> > +
>
> Should we break here if nr_to_scan is zero?
Not a very likely scenario, but ok.
>
> > + if (shrinker->flags & SHRINKER_MEMCG_AWARE) {
> > + memcg = mem_cgroup_get_from_ino(ino);
> > + if (!memcg || IS_ERR(memcg))
>
> Should we drop the check of "!memcg" since mem_cgroup_get_from_ino
> cannot return NULL?
It can if !CONFIG_MEMCG. You might argue that then shrinker can not have
the SHRINKER_MEMCG_AWARE flag, but since it's not a hot path at all,
I'll keep it for extra safety.
>
> > + return -ENOENT;
> > +
> > + if (!mem_cgroup_online(memcg)) {
> > + mem_cgroup_put(memcg);
> > + return -ENOENT;
> > + }
> > + } else {
> > + if (ino != 0)
> > + return -EINVAL;
> > + memcg = NULL;
>
> IIUC, memcg is already NULL if we reach here, right? Then the
> assignment is not necessary. Or we cound remove the initialization
> of 'memcg' where it is definned.
Right, removed.
>
> > + }
> > +
> > + ret = down_read_killable(&shrinker_rwsem);
> > + if (ret) {
> > + mem_cgroup_put(memcg);
> > + return ret;
> > + }
> > +
> > + sc.nid = nid;
> > + sc.memcg = memcg;
> > + sc.nr_to_scan = nr_to_scan;
> > + sc.nr_scanned = nr_to_scan;
> > +
> > + shrinker->scan_objects(shrinker, &sc);
> > +
> > + up_read(&shrinker_rwsem);
> > + mem_cgroup_put(memcg);
> > +
> > + return ret ? ret : size;
>
> Seems "ret" is always equal to 0 here, should we simplify this
> to "return size"?
Right.
Thank you for the review!
