DTrace fixes for node.js

[Mark Johnston]

On Mon, Dec 01, 2014 at 04:59:02PM +0400, Fedor Indutny wrote:
> Hello Mark, DTrace people!
> 

Hi Fedor,

> Hope you are doing well.
> 
> Looks like FreeBSD 10.1 decided to play some not that funny games
> with drti.c again :)
> 
> I have submitted a [patch][0] as a problem report.

Thank you. I'll test and commit it if I don't run into any problems. 

> 
> It fixes iteration over aligned DOFs, and accounts for a
> non-master-branch FreeBSD stuff in:
> 
> cddl/contrib/opensolaris/lib/libdtrace/common/dt_link.c
> 
> Which is currently producing SHT_PROGBITS section instead of
> expected SHT_SUNW_dof, thus confusing the drti.c file.

Hm, are you referring to this change:
https://svnweb.freebsd.org/base?view=revision&revision=271413
?

I'm a bit confused, as it hasn't been merged to 10 yet: it exposed
another dtrace bug that caused build failures in some USDT-enabled
ports, so I'm waiting for the subsequent fixes to settle.

-Mark

> 
> Thank you,
> Fedor.
> 
> [0]: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=195555
> 
> 
> On Thu, Feb 27, 2014 at 6:50 PM, Fedor Indutny <fedor at indutny.com> wrote:
> 
> > Update sent, thank you!
> >
> > On Thu, Feb 27, 2014 at 6:39 PM, Mark Johnston <markj at freebsd.org> wrote:
> > > On Thu, Feb 27, 2014 at 3:30 AM, Fedor Indutny <fedor at indutny.com>
> > wrote:
> > >> Mark,
> > >>
> > >> Thanks for looking into this. I just tried your patch and it (no
> > >> surprise) builds fine. Node.js DOF symbols seems to be resolving
> > >> properly too!
> > >>
> > >> Do you want me to squash this changes into my patch, and post them on
> > >> that ticket?
> > >>
> > >
> > > That would be good, thanks. When I have some time I'll do more testing
> > > and commit the change.
> > >
> > > -Mark
> > >
> > >> On Thu, Feb 27, 2014 at 9:01 AM, Mark Johnston <markj at freebsd.org>
> > wrote:
> > >>> On Tue, Feb 25, 2014 at 06:16:15PM +0400, Fedor Indutny wrote:
> > >>>> Hello devs!
> > >>>>
> > >>>> I have made some fixes to fix DTrace support for node.js in FreeBSD:
> > >>>>
> > >>>> * http://www.freebsd.org/cgi/query-pr.cgi?pr=186821
> > >>>> * http://www.freebsd.org/cgi/query-pr.cgi?pr=187027
> > >>>>
> > >>>> Here is a blog post with a bit of explanation of why this is needed
> > >>>> and what is fixed: https://blog.indutny.com/7.freebsd-dtrace
> > >>>>
> > >>>> Please let me know if I could be any help in reviewing it.
> > >>>
> > >>> Hi Fedor,
> > >>>
> > >>> The DOF limit change looks fine to me. I note that the illumos guys
> > have
> > >>> just pushed a change to illumos-gate which bumps dtrace_dof_maxsize,
> > but
> > >>> it's good to have the sysctl as well.
> > >>>
> > >>> The drti change looks mostly good to me. The real problem there is that
> > >>> our linker doesn't know how to merge DOF, so it just concatenates the
> > >>> tables into one SUNW_dof section. So we should really fix our linker,
> > >>> but it doesn't hurt to also handle the problem in drti.o.
> > >>>
> > >>> There are a couple of bugs in the patch. First, the "break" added after
> > >>> finding the DOF section causes problems if we haven't yet seen the
> > >>> symbol table. Second, fixedprobes needs to be reset at the beginning of
> > >>> each iteration of the while loop that you added, else we may not try
> > >>> searching the dynamic symbol table when fixing the probe addresses.
> > I've
> > >>> pasted a patch below; could you test it and make sure things still work
> > >>> properly with node?
> > >>>
> > >>> Thanks for the detailed blog post and problem description, they were
> > >>> very helpful. :)
> > >>>
> > >>> -Mark
> > >>>
> > >>> diff --git a/cddl/contrib/opensolaris/lib/libdtrace/common/drti.c
> > b/cddl/contrib/opensolaris/lib/libdtrace/common/drti.c
> > >>> index e47cfb4d..bb02d8c 100644
> > >>> --- a/cddl/contrib/opensolaris/lib/libdtrace/common/drti.c
> > >>> +++ b/cddl/contrib/opensolaris/lib/libdtrace/common/drti.c
> > >>> @@ -162,7 +162,7 @@ dtrace_dof_init(void)
> > >>>         char *dofstrtabraw;
> > >>>         size_t shstridx, symtabidx = 0, dynsymidx = 0;
> > >>>         unsigned char *buf;
> > >>> -       int fixedprobes = 0;
> > >>> +       int fixedprobes;
> > >>>  #endif
> > >>>
> > >>>         if (getenv("DTRACE_DOF_INIT_DISABLE") != NULL)
> > >>> @@ -214,7 +214,6 @@ dtrace_dof_init(void)
> > >>>                         if  (s && strcmp(s, ".SUNW_dof") == 0) {
> > >>>                                 dofdata = elf_getdata(scn, NULL);
> > >>>                                 dof = dofdata->d_buf;
> > >>> -                               break;
> > >>>                         }
> > >>>                 }
> > >>>         }
> > >>> @@ -226,6 +225,7 @@ dtrace_dof_init(void)
> > >>>         }
> > >>>
> > >>>         while ((char *) dof < (char *) dofdata->d_buf +
> > dofdata->d_size) {
> > >>> +               fixedprobes = 0;
> > >>>                 dof_next = (void *) ((char *) dof + dof->dofh_filesz);
> > >>>  #endif
> > >>>
> >