Re: [PATCH v2] mm: emit tracepoint when RSS changes by threshold

[Joel Fernandes]

On Tue, Sep 03, 2019 at 09:44:51PM -0700, Suren Baghdasaryan wrote:
> On Tue, Sep 3, 2019 at 1:09 PM Joel Fernandes (Google)
> <joel@xxxxxxxxxxxxxxxxx> wrote:
> >
> > Useful to track how RSS is changing per TGID to detect spikes in RSS and
> > memory hogs. Several Android teams have been using this patch in various
> > kernel trees for half a year now. Many reported to me it is really
> > useful so I'm posting it upstream.
> >
> > Initial patch developed by Tim Murray. Changes I made from original patch:
> > o Prevent any additional space consumed by mm_struct.
> > o Keep overhead low by checking if tracing is enabled.
> > o Add some noise reduction and lower overhead by emitting only on
> >   threshold changes.
> >
> > Co-developed-by: Tim Murray <timmurray@xxxxxxxxxx>
> > Signed-off-by: Tim Murray <timmurray@xxxxxxxxxx>
> > Signed-off-by: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> >
> > ---
> >
> > v1->v2: Added more commit message.
> >
> > Cc: carmenjackson@xxxxxxxxxx
> > Cc: mayankgupta@xxxxxxxxxx
> > Cc: dancol@xxxxxxxxxx
> > Cc: rostedt@xxxxxxxxxxx
> > Cc: minchan@xxxxxxxxxx
> > Cc: akpm@xxxxxxxxxxxxxxxxxxxx
> > Cc: kernel-team@xxxxxxxxxxx
> >
> >  include/linux/mm.h          | 14 +++++++++++---
> >  include/trace/events/kmem.h | 21 +++++++++++++++++++++
> >  mm/memory.c                 | 20 ++++++++++++++++++++
> >  3 files changed, 52 insertions(+), 3 deletions(-)
> >
> > diff --git a/include/linux/mm.h b/include/linux/mm.h
> > index 0334ca97c584..823aaf759bdb 100644
> > --- a/include/linux/mm.h
> > +++ b/include/linux/mm.h
> > @@ -1671,19 +1671,27 @@ static inline unsigned long get_mm_counter(struct mm_struct *mm, int member)
> >         return (unsigned long)val;
> >  }
> >
> > +void mm_trace_rss_stat(int member, long count, long value);
> > +
> >  static inline void add_mm_counter(struct mm_struct *mm, int member, long value)
> >  {
> > -       atomic_long_add(value, &mm->rss_stat.count[member]);
> > +       long count = atomic_long_add_return(value, &mm->rss_stat.count[member]);
> > +
> > +       mm_trace_rss_stat(member, count, value);
> >  }
> >
> >  static inline void inc_mm_counter(struct mm_struct *mm, int member)
> >  {
> > -       atomic_long_inc(&mm->rss_stat.count[member]);
> > +       long count = atomic_long_inc_return(&mm->rss_stat.count[member]);
> > +
> > +       mm_trace_rss_stat(member, count, 1);
> >  }
> >
> >  static inline void dec_mm_counter(struct mm_struct *mm, int member)
> >  {
> > -       atomic_long_dec(&mm->rss_stat.count[member]);
> > +       long count = atomic_long_dec_return(&mm->rss_stat.count[member]);
> > +
> > +       mm_trace_rss_stat(member, count, -1);
> >  }
> >
> >  /* Optimized variant when page is already known not to be PageAnon */
> > diff --git a/include/trace/events/kmem.h b/include/trace/events/kmem.h
> > index eb57e3037deb..8b88e04fafbf 100644
> > --- a/include/trace/events/kmem.h
> > +++ b/include/trace/events/kmem.h
> > @@ -315,6 +315,27 @@ TRACE_EVENT(mm_page_alloc_extfrag,
> >                 __entry->change_ownership)
> >  );
> >
> > +TRACE_EVENT(rss_stat,
> > +
> > +       TP_PROTO(int member,
> > +               long count),
> > +
> > +       TP_ARGS(member, count),
> > +
> > +       TP_STRUCT__entry(
> > +               __field(int, member)
> > +               __field(long, size)
> > +       ),
> > +
> > +       TP_fast_assign(
> > +               __entry->member = member;
> > +               __entry->size = (count << PAGE_SHIFT);
> > +       ),
> > +
> > +       TP_printk("member=%d size=%ldB",
> > +               __entry->member,
> > +               __entry->size)
> > +       );
> >  #endif /* _TRACE_KMEM_H */
> >
> >  /* This part must be outside protection */
> > diff --git a/mm/memory.c b/mm/memory.c
> > index e2bb51b6242e..9d81322c24a3 100644
> > --- a/mm/memory.c
> > +++ b/mm/memory.c
> > @@ -72,6 +72,8 @@
> >  #include <linux/oom.h>
> >  #include <linux/numa.h>
> >
> > +#include <trace/events/kmem.h>
> > +
> >  #include <asm/io.h>
> >  #include <asm/mmu_context.h>
> >  #include <asm/pgalloc.h>
> > @@ -140,6 +142,24 @@ static int __init init_zero_pfn(void)
> >  }
> >  core_initcall(init_zero_pfn);
> >
> > +/*
> > + * This threshold is the boundary in the value space, that the counter has to
> > + * advance before we trace it. Should be a power of 2. It is to reduce unwanted
> > + * trace overhead. The counter is in units of number of pages.
> > + */
> > +#define TRACE_MM_COUNTER_THRESHOLD 128
> 
> IIUC the counter has to change by 128 pages (512kB assuming 4kB pages)
> before the change gets traced. Would it make sense to make this step
> size configurable? For a system with limited memory size change of
> 512kB might be considerable while on systems with plenty of memory
> that might be negligible. Not even mentioning possible difference in
> page sizes. Maybe something like
> /sys/kernel/debug/tracing/rss_step_order with
> TRACE_MM_COUNTER_THRESHOLD=(1<<rss_step_order)?

I would not want to complicate this more to be honest. It is already a bit
complex, and I am not sure about the win in making it as configurable as you
seem to want. The "threshold" thing is just a slight improvement, it is not
aiming to be optimal. If in your tracing, this granularity is an issue, we
can visit it then.

thanks,

 - Joel



> > +void mm_trace_rss_stat(int member, long count, long value)
> > +{
> > +       long thresh_mask = ~(TRACE_MM_COUNTER_THRESHOLD - 1);
> > +
> > +       if (!trace_rss_stat_enabled())
> > +               return;
> > +
> > +       /* Threshold roll-over, trace it */
> > +       if ((count & thresh_mask) != ((count - value) & thresh_mask))
> > +               trace_rss_stat(member, count);
> > +}
> >
> >  #if defined(SPLIT_RSS_COUNTING)
> >
> > --
> > 2.23.0.187.g17f5b7556c-goog
> >
> > --
> > To unsubscribe from this group and stop receiving emails from it, send an email to kernel-team+unsubscribe@xxxxxxxxxxxx
> >