Linux-mm Archive on lore.kernel.org
 help / color / Atom feed
From: Yafang Shao <laoar.shao@gmail.com>
To: Yang Shi <shy828301@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	Michal Hocko <mhocko@suse.com>,  Linux MM <linux-mm@kvack.org>,
	Yafang Shao <shaoyafang@didiglobal.com>,
	 Johannes Weiner <hannes@cmpxchg.org>
Subject: Re: [PATCH] mm/vmscan: expose cgroup_ino for memcg reclaim tracepoints
Date: Sat, 13 Jul 2019 14:46:47 +0800
Message-ID: <CALOAHbAq_y2Dng9xxWe4NdNrHa35CpQkrjiYjYHck2Hz8jGssQ@mail.gmail.com> (raw)
In-Reply-To: <CAHbLzkqw+LJC-CrpJpZBfoer9jNRAcfZz+YTLP1qqa_x7R8y1w@mail.gmail.com>

On Sat, Jul 13, 2019 at 7:58 AM Yang Shi <shy828301@gmail.com> wrote:
>
> On Thu, Jul 11, 2019 at 6:10 PM Andrew Morton <akpm@linux-foundation.org> wrote:
> >
> >
> > Can we please get some review of this one?  It has been in -mm since
> > May 22, no issues that I've heard of.
> >
> >
> > From: Yafang Shao <laoar.shao@gmail.com>
> > Subject: mm/vmscan: expose cgroup_ino for memcg reclaim tracepoints
> >
> > We can use the exposed cgroup_ino to trace specified cgroup.
> >
> > For example,
> > step 1, get the inode of the specified cgroup
> >         $ ls -di /tmp/cgroupv2/foo
> > step 2, set this inode into tracepoint filter to trace this cgroup only
> >         (assume the inode is 11)
> >         $ cd /sys/kernel/debug/tracing/events/vmscan/
> >         $ echo 'cgroup_ino == 11' > mm_vmscan_memcg_reclaim_begin/filter
> >         $ echo 'cgroup_ino == 11' > mm_vmscan_memcg_reclaim_end/filter
> >
> > The reason I made this change is to trace a specific container.
>
> I'm wondering how useful this is. You could filter events by cgroup
> with bpftrace easily. For example:
>
> # bpftrace -e 'tracepoint:syscalls:sys_enter_openat /cgroup ==
> cgroupid("/sys/fs/cgroup/unified/mycg")/ { printf("%s\n",
> str(args->filename)); }':
>

Seems the bpftrace get the cgroupid from the current task and then
compare the task-cgroupid with the speficied cgroupid?
While in the memcg reclaim, the pages in a memcg may be reclaimed by a
process in other memcgs, i.e. the parent memcg,
so we can't use the process's memcg as the filter.

The way to use bpftrace here is using kprobe to do it, I guess.
But as the tracepoint is already there, we can make little change to enhance it.

Thanks
Yafang

>
> >
> > Sometimes there're lots of containers on one host.  Some of them are
> > not important at all, so we don't care whether them are under memory
> > pressure.  While some of them are important, so we want't to know if
> > these containers are doing memcg reclaim and how long this relaim
> > takes.
> >
> > Without this change, we don't know the memcg reclaim happend in which
> > container.
> >
> > Link: http://lkml.kernel.org/r/1557649528-11676-1-git-send-email-laoar.shao@gmail.com
> > Signed-off-by: Yafang Shao <laoar.shao@gmail.com>
> > Cc: Michal Hocko <mhocko@suse.com>
> > Cc: <shaoyafang@didiglobal.com>
> > Cc: Johannes Weiner <hannes@cmpxchg.org>
> > Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
> > ---
> >
> >  include/trace/events/vmscan.h |   71 ++++++++++++++++++++++++++------
> >  mm/vmscan.c                   |   18 +++++---
> >  2 files changed, 72 insertions(+), 17 deletions(-)
> >
> > --- a/include/trace/events/vmscan.h~mm-vmscan-expose-cgroup_ino-for-memcg-reclaim-tracepoints
> > +++ a/include/trace/events/vmscan.h
> > @@ -127,18 +127,43 @@ DEFINE_EVENT(mm_vmscan_direct_reclaim_be
> >  );
> >
> >  #ifdef CONFIG_MEMCG
> > -DEFINE_EVENT(mm_vmscan_direct_reclaim_begin_template, mm_vmscan_memcg_reclaim_begin,
> > +DECLARE_EVENT_CLASS(mm_vmscan_memcg_reclaim_begin_template,
> >
> > -       TP_PROTO(int order, gfp_t gfp_flags),
> > +       TP_PROTO(unsigned int cgroup_ino, int order, gfp_t gfp_flags),
> >
> > -       TP_ARGS(order, gfp_flags)
> > +       TP_ARGS(cgroup_ino, order, gfp_flags),
> > +
> > +       TP_STRUCT__entry(
> > +               __field(unsigned int, cgroup_ino)
> > +               __field(int, order)
> > +               __field(gfp_t, gfp_flags)
> > +       ),
> > +
> > +       TP_fast_assign(
> > +               __entry->cgroup_ino     = cgroup_ino;
> > +               __entry->order          = order;
> > +               __entry->gfp_flags      = gfp_flags;
> > +       ),
> > +
> > +       TP_printk("cgroup_ino=%u order=%d gfp_flags=%s",
> > +               __entry->cgroup_ino, __entry->order,
> > +               show_gfp_flags(__entry->gfp_flags))
> >  );
> >
> > -DEFINE_EVENT(mm_vmscan_direct_reclaim_begin_template, mm_vmscan_memcg_softlimit_reclaim_begin,
> > +DEFINE_EVENT(mm_vmscan_memcg_reclaim_begin_template,
> > +            mm_vmscan_memcg_reclaim_begin,
> >
> > -       TP_PROTO(int order, gfp_t gfp_flags),
> > +       TP_PROTO(unsigned int cgroup_ino, int order, gfp_t gfp_flags),
> >
> > -       TP_ARGS(order, gfp_flags)
> > +       TP_ARGS(cgroup_ino, order, gfp_flags)
> > +);
> > +
> > +DEFINE_EVENT(mm_vmscan_memcg_reclaim_begin_template,
> > +            mm_vmscan_memcg_softlimit_reclaim_begin,
> > +
> > +       TP_PROTO(unsigned int cgroup_ino, int order, gfp_t gfp_flags),
> > +
> > +       TP_ARGS(cgroup_ino, order, gfp_flags)
> >  );
> >  #endif /* CONFIG_MEMCG */
> >
> > @@ -167,18 +192,40 @@ DEFINE_EVENT(mm_vmscan_direct_reclaim_en
> >  );
> >
> >  #ifdef CONFIG_MEMCG
> > -DEFINE_EVENT(mm_vmscan_direct_reclaim_end_template, mm_vmscan_memcg_reclaim_end,
> > +DECLARE_EVENT_CLASS(mm_vmscan_memcg_reclaim_end_template,
> >
> > -       TP_PROTO(unsigned long nr_reclaimed),
> > +       TP_PROTO(unsigned int cgroup_ino, unsigned long nr_reclaimed),
> >
> > -       TP_ARGS(nr_reclaimed)
> > +       TP_ARGS(cgroup_ino, nr_reclaimed),
> > +
> > +       TP_STRUCT__entry(
> > +               __field(unsigned int, cgroup_ino)
> > +               __field(unsigned long, nr_reclaimed)
> > +       ),
> > +
> > +       TP_fast_assign(
> > +               __entry->cgroup_ino     = cgroup_ino;
> > +               __entry->nr_reclaimed   = nr_reclaimed;
> > +       ),
> > +
> > +       TP_printk("cgroup_ino=%u nr_reclaimed=%lu",
> > +               __entry->cgroup_ino, __entry->nr_reclaimed)
> >  );
> >
> > -DEFINE_EVENT(mm_vmscan_direct_reclaim_end_template, mm_vmscan_memcg_softlimit_reclaim_end,
> > +DEFINE_EVENT(mm_vmscan_memcg_reclaim_end_template,
> > +            mm_vmscan_memcg_reclaim_end,
> >
> > -       TP_PROTO(unsigned long nr_reclaimed),
> > +       TP_PROTO(unsigned int cgroup_ino, unsigned long nr_reclaimed),
> >
> > -       TP_ARGS(nr_reclaimed)
> > +       TP_ARGS(cgroup_ino, nr_reclaimed)
> > +);
> > +
> > +DEFINE_EVENT(mm_vmscan_memcg_reclaim_end_template,
> > +            mm_vmscan_memcg_softlimit_reclaim_end,
> > +
> > +       TP_PROTO(unsigned int cgroup_ino, unsigned long nr_reclaimed),
> > +
> > +       TP_ARGS(cgroup_ino, nr_reclaimed)
> >  );
> >  #endif /* CONFIG_MEMCG */
> >
> > --- a/mm/vmscan.c~mm-vmscan-expose-cgroup_ino-for-memcg-reclaim-tracepoints
> > +++ a/mm/vmscan.c
> > @@ -3191,8 +3191,10 @@ unsigned long mem_cgroup_shrink_node(str
> >         sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) |
> >                         (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK);
> >
> > -       trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order,
> > -                                                     sc.gfp_mask);
> > +       trace_mm_vmscan_memcg_softlimit_reclaim_begin(
> > +                                       cgroup_ino(memcg->css.cgroup),
> > +                                       sc.order,
> > +                                       sc.gfp_mask);
> >
> >         /*
> >          * NOTE: Although we can get the priority field, using it
> > @@ -3203,7 +3205,9 @@ unsigned long mem_cgroup_shrink_node(str
> >          */
> >         shrink_node_memcg(pgdat, memcg, &sc, &lru_pages);
> >
> > -       trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed);
> > +       trace_mm_vmscan_memcg_softlimit_reclaim_end(
> > +                                       cgroup_ino(memcg->css.cgroup),
> > +                                       sc.nr_reclaimed);
> >
> >         *nr_scanned = sc.nr_scanned;
> >         return sc.nr_reclaimed;
> > @@ -3241,7 +3245,9 @@ unsigned long try_to_free_mem_cgroup_pag
> >
> >         zonelist = &NODE_DATA(nid)->node_zonelists[ZONELIST_FALLBACK];
> >
> > -       trace_mm_vmscan_memcg_reclaim_begin(0, sc.gfp_mask);
> > +       trace_mm_vmscan_memcg_reclaim_begin(
> > +                               cgroup_ino(memcg->css.cgroup),
> > +                               0, sc.gfp_mask);
> >
> >         psi_memstall_enter(&pflags);
> >         noreclaim_flag = memalloc_noreclaim_save();
> > @@ -3251,7 +3257,9 @@ unsigned long try_to_free_mem_cgroup_pag
> >         memalloc_noreclaim_restore(noreclaim_flag);
> >         psi_memstall_leave(&pflags);
> >
> > -       trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed);
> > +       trace_mm_vmscan_memcg_reclaim_end(
> > +                               cgroup_ino(memcg->css.cgroup),
> > +                               nr_reclaimed);
> >
> >         return nr_reclaimed;
> >  }
> > _
> >


      reply index

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-04-09 10:11 Yafang Shao
2019-05-10 19:24 ` Dmitry Dolgov
2019-05-11  0:51   ` Yafang Shao
2019-07-12  1:10 ` Andrew Morton
2019-07-12 23:58   ` Yang Shi
2019-07-13  6:46     ` Yafang Shao [this message]

Reply instructions:

You may reply publically 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=CALOAHbAq_y2Dng9xxWe4NdNrHa35CpQkrjiYjYHck2Hz8jGssQ@mail.gmail.com \
    --to=laoar.shao@gmail.com \
    --cc=akpm@linux-foundation.org \
    --cc=hannes@cmpxchg.org \
    --cc=linux-mm@kvack.org \
    --cc=mhocko@suse.com \
    --cc=shaoyafang@didiglobal.com \
    --cc=shy828301@gmail.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

Linux-mm Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-mm/0 linux-mm/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-mm linux-mm/ https://lore.kernel.org/linux-mm \
		linux-mm@kvack.org linux-mm@archiver.kernel.org
	public-inbox-index linux-mm


Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kvack.linux-mm


AGPL code for this site: git clone https://public-inbox.org/ public-inbox