Re: opjitconv runs indefinitely

[大平怜 <rei...@gm...>]

Thanks much, Will!

I confirmed the latest oprofile in git solved the problem for OpenJDK on
ppc Linux.


Rei Odaira

2015-05-22 14:43 GMT-05:00 William Cohen <wc...@re...>:

> On 05/05/2015 02:50 PM, 大平怜 wrote:
> > Hi,
> >
> > I found opjitconv ran indefinitely when profiling a Java application
> running on
> > OpenJDK/ppc64le.  This is because OpenJDK sometimes reports generation of
> > zero-size jitted code via JVMTI, but scan_overlaps() in opjitconv does
> not assume
> > the existence of jitted code with size zero.
> >
> > (1) scan_overlaps() finds overlap between a normal jitted code and a
> zero-size
> >      jitted code.
> > (2) eliminate_overlaps() tries to split the zero-size jitted code but
> cannot.
> > (3) resolve_overlaps() incorrectly thinks the split has happened and
> invokes
> >      scan_overlaps() again.
> > (4) Back to (1)
> >
> > One workaround would be to remove all the zero-size entries before
> resolving
> > overlaps (patch attached), but I am not sure if this is a good solution.
> > It works at least in my environment.
> >
> >
> > Regards,
> > Rei Odaira
>
> Hi Rei,
>
> Thanks for the patch. I corrected the formatting of the patch and checked
> it in.  It is now in the upstream oprofile git repository.  Could you
> verify that the upstream oprofile solves the observed problem?
>
> -Will
> >
> >
> > --- oprofile-1.0.0/opjitconv/jitsymbol.c2014-09-12 09:39:47.000000000
> -0500
> > +++ oprofile-1.0.0-openjdk8-src/opjitconv/jitsymbol.c2015-05-05
> 13:37:13.931202002 -0500
> > @@ -201,6 +201,26 @@
> >  }
> >  }
> >
> > +static void invalidate_zero_size_entries(void)
> > +{
> > +u32 i;
> > +int flag;
> > +struct jitentry * a;
> > +
> > +flag = 0;
> > +for (i = 0; i < entry_count; i++) {
> > +a = entries_address_ascending[i];
> > +if (a->code_size == 0) {
> > +invalidate_entry(a);
> > +flag = 1;
> > +}
> > +}
> > +if (flag) {
> > +resort_address();
> > +resort_symbol();
> > +}
> > +}
> > +
> >
> >  /* select the symbol with the longest life time in the index range */
> >  static int select_one(int start_idx, int end_idx)
> > @@ -505,6 +525,7 @@
> >  int cnt = 0;
> >
> >  invalidate_earlybirds(start_time);
> > +invalidate_zero_size_entries();
> >  while ((rc = scan_overlaps()) && rc != OP_JIT_CONV_FAIL) {
> >  resort_address();
> >  if (cnt == 0) {
> >
> >
> >
> >
> ------------------------------------------------------------------------------
> > One dashboard for servers and applications across Physical-Virtual-Cloud
> > Widest out-of-the-box monitoring support with 50+ applications
> > Performance metrics, stats and reports that give you Actionable Insights
> > Deep dive visibility with transaction tracing using APM Insight.
> > http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
> >
> >
> >
> > _______________________________________________
> > oprofile-list mailing list
> > opr...@li...
> > https://lists.sourceforge.net/lists/listinfo/oprofile-list
> >
>
>