Re: [patch 026/158] mm: emit tracepoint when RSS changes

[Joel Fernandes <joel@joelfernandes.org>]

On Mon, Dec 02, 2019 at 04:56:01PM -0500, Steven Rostedt wrote:
> On Mon, 2 Dec 2019 16:13:45 -0500
> Joel Fernandes <joel@joelfernandes.org> wrote:
> 
> > On Mon, Dec 02, 2019 at 12:14:15PM -0500, Steven Rostedt wrote:
> > > On Sat, 30 Nov 2019 17:50:30 -0800
> > > akpm@linux-foundation.org wrote:
> > >   
> > > >  /* Optimized variant when page is already known not to be PageAnon */
> > > > --- a/include/trace/events/kmem.h~mm-emit-tracepoint-when-rss-changes
> > > > +++ a/include/trace/events/kmem.h
> > > > @@ -316,6 +316,27 @@ 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);  
> > > 
> > > It's best to put all calculations (including shifts) in the print part,
> > > as that's the slow path. The TP_fast_assign() is done when the trace
> > > point is triggered (during the execution of the code). It's best to
> > > keep this in the slow path (TP_printk).
> > > 
> > > 		__entry->count = count;
> > >   
> > > > +	),
> > > > +
> > > > +	TP_printk("member=%d size=%ldB",
> > > > +		__entry->member,
> > > > +		__entry->size)  
> > > 
> > > 		__entry->count << PAGE_SHIFT)  
> > 
> > Ah. Android users now use a tool called perfetto which gather raw trace
> > (binary format). So such shifting will have to be done by userspace then if
> > we did it this way. And I'm afraid this patch has been in circulation for
> > quite some time now that may cause major pains in changing userspace tooling
> > now :-\
> > 
> 
> I've been trying to get libtraceevent into a standard library, and
> it's been ready to go. Can your tool please use that. It can handle this
> without issue. Just feed it the format file (that's exported to
> user space via the tracefs file system).

I would love for that to happen but I don't develop Perfetto much. If I am
writing a tool I will definitely give it a go from my side. CC'ing Perfetto's
lead developer Primiano -- I believe you have already met Primiano at a
conference before as he mentioned it to me that you guys met. I also believe
this topic of using a common library was discussed before, but something
about licensing came up.

thanks,

 - Joel