On Mon, Aug 10, 2009 at 12:22:12AM -0700, Manfred M?cke wrote:
> > If dtrace thinks something fishy is going on, it can refuse to instrument
> > the return sites. This is mainly due to issues with jump-tables.
> > If you set DTRACE_DEBUG= in your environment, is there anything of interest?
> > Look for "grabbed pid NNNN" as a starting place.
>
> Thanks for the hint, I was not aware of DTRACE_DEBUG. I found the following,
> which confirms your assumption, that it might be an issue with jump-tables:
>
> libdtrace DEBUG: creating probe pid17864:myprog:dcntrl_:return
> libdtrace DEBUG: error at 0x851 (assuming jump table)
>
> I searched the forum but haven't found an explanation under what
> conditions dtrace assumes a jump-table. Are there any (Fortran)
> compiler settings which could help avoiding this issue?
That error is from usr/src/lib/libdtrace/i386/dt_isadep.c:
78 dt_pid_has_jump_table(struct ps_prochandle *P, dtrace_hdl_t *dtp,
79 uint8_t *text, fasttrap_probe_spec_t *ftp, const GElf_Sym *symp)
80 {
...
91 for (i = 0; i < ftp->ftps_size; i += size) {
92 size = dt_instr_size(&text[i], dtp, pid, symp->st_value
+ i,
93 dmodel);
95 /*
96 * Assume the worst if we hit an illegal instruction.
97 */
98 if (size <= 0) {
99 dt_dprintf("error at %#lx (assuming jump
table)\n", i);
100 return (1);
101 }
...
115 }
116
117 return (0);
118 }
This means that there is some instruction in the stream that dtrace's
disassembler is either getting the size wrong on (causing us to get
out of alignment, causing us at a later point to fail on a bad instruction),
or we're hitting an instruction we don't understand.
Could you send me the output of:
"echo dcntrl_::dis | mdb ./myprog"
which uses the same disassembler.
Cheers,
- jonathan
> In what context should I read the given address 0x851? Neither the
> function dcntrl_ nor the calling function has an opcode starting at
> this address.
It's an offset
_______________________________________________
dtrace-discuss mailing list
[email protected]
