Re: [PATCH v1 1/2] mm: memcg: print out cgroup name in the memcg tracepoints

[Dmitry Rokosov]

Hello Andrii!

Thank you for quick feedback!

On Wed, Nov 01, 2023 at 10:08:34AM -0700, Andrii Nakryiko wrote:
> On Wed, Nov 1, 2023 at 3:29 AM Dmitry Rokosov
> <ddrokosov@xxxxxxxxxxxxxxxxx> wrote:
> >
> > Sometimes it is necessary to understand in which memcg tracepoint event
> > occurred. The function cgroup_name() is a useful tool for this purpose.
> > To integrate cgroup_name() into the existing memcg tracepoints, this
> > patch introduces a new tracepoint template for the begin() and end()
> > events, utilizing static __array() to store the cgroup name.
> >
> > Signed-off-by: Dmitry Rokosov <ddrokosov@xxxxxxxxxxxxxxxxx>
> > ---
> >  include/trace/events/vmscan.h | 77 +++++++++++++++++++++++++++++------
> >  mm/vmscan.c                   |  8 ++--
> >  2 files changed, 69 insertions(+), 16 deletions(-)
> >
> > diff --git a/include/trace/events/vmscan.h b/include/trace/events/vmscan.h
> > index d2123dd960d5..124bc22866c8 100644
> > --- a/include/trace/events/vmscan.h
> > +++ b/include/trace/events/vmscan.h
> > @@ -141,19 +141,47 @@ DEFINE_EVENT(mm_vmscan_direct_reclaim_begin_template, mm_vmscan_direct_reclaim_b
> >  );
> >
> >  #ifdef CONFIG_MEMCG
> > -DEFINE_EVENT(mm_vmscan_direct_reclaim_begin_template, mm_vmscan_memcg_reclaim_begin,
> >
> > -       TP_PROTO(int order, gfp_t gfp_flags),
> > +DECLARE_EVENT_CLASS(mm_vmscan_memcg_reclaim_begin_template,
> >
> > -       TP_ARGS(order, gfp_flags)
> > +       TP_PROTO(const struct mem_cgroup *memcg, int order, gfp_t gfp_flags),
> > +
> > +       TP_ARGS(memcg, order, gfp_flags),
> 
> By adding memcg in front of existing tracepoint arguments, you
> unnecessarily break everyone who currently has some scripts based on
> this tracepoint. Given there is no reason why memcg has to be the very
> first argument, it would be nice if you can just append it at the end
> to make it nicely backwards compatible. Same for other tracepoints
> below.
> 
> Tracepoints are not an ABI, but there is also no point in arbitrarily
> breaking all current scripts for such a trivial reason.
> 

You are absolutely correct. I didn't consider the scripts that rely on
these tracepoints, because tracepoints are not an ABI, as you mentioned.
Additionally, I added the memcg parameter as the first argument based on
my personal programming patterns, where the context object should always
come first :)

I apologize for that and will prepare new version.

> > +
> > +       TP_STRUCT__entry(
> > +               __field(int, order)
> > +               __field(unsigned long, gfp_flags)
> > +               __array(char, name, NAME_MAX + 1)
> > +       ),
> > +
> > +       TP_fast_assign(
> > +               __entry->order = order;
> > +               __entry->gfp_flags = (__force unsigned long)gfp_flags;
> > +               cgroup_name(memcg->css.cgroup,
> > +                       __entry->name,
> > +                       sizeof(__entry->name));
> > +       ),
> > +
> > +       TP_printk("memcg=%s order=%d gfp_flags=%s",
> > +               __entry->name,
> > +               __entry->order,
> > +               show_gfp_flags(__entry->gfp_flags))
> >  );
> 
> [...]

-- 
Thank you,
Dmitry