DTrace fixes for node.js
Mark Johnston
markj at freebsd.org
Tue Dec 2 00:13:48 UTC 2014
On Tue, Dec 02, 2014 at 03:02:19AM +0400, Fedor Indutny wrote:
> Mark,
>
> Yes, 271413 is the one fixing PROGBITS. Anyway, my patch should
> support both PROGBITS and DOFW_sun, making it quite universal.
But why is it necessary? As of r271413, the DOF section type should
always be SUNW_dof. If not, that's unexpected, and I'd like to understand
what's going on.
On 10.1, we're still creating a section of type PROGBITS, and that's
what drti.o is looking for. Is there a regression?
Thanks,
-Mark
>
> Thank you for your time!
>
> Fedor.
>
> On Tue, Dec 2, 2014 at 12:54 AM, Mark Johnston <markj at freebsd.org> wrote:
>
> > 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
> > > > >>>
> > > >
> >
More information about the freebsd-dtrace
mailing list