Re: The newest gdb 8.3 crash on OpenBSD

[Nan Xiao <xiaonan830818@gmail.com>]

Hi Simon,

Thanks very much for your reply!

Yes, you are correct! After checking gdb/config.h, I found all ptrace
related macros are undefined:

/* Define if sys/ptrace.h defines the PTRACE_GETFPXREGS request. */
/* #undef HAVE_PTRACE_GETFPXREGS */

/* Define if sys/ptrace.h defines the PTRACE_GETREGS request. */
/* #undef HAVE_PTRACE_GETREGS */

/* Define to 1 if you have the <ptrace.h> header file. */
/* #undef HAVE_PTRACE_H */

/* Define if sys/ptrace.h defines the PT_GETDBREGS request. */
/* #undef HAVE_PT_GETDBREGS */

/* Define if sys/ptrace.h defines the PT_GETXMMREGS request. */
/* #undef HAVE_PT_GETXMMREGS */

I checked the header files, there is no /usr/include/ptrace.h. There
is /usr/include/sys/ptrace.h, but the real definitions (such as
PT_GETREGS, PT_SETREGS, etc) are in system related header files. E.g.,
in my OS, it is in /usr/include/amd64/ptrace.h. I checked OpenBSD's
gdb port (gdb 7.12.1, http://openports.se/devel/gdb), it seems I need
more tweak to make gdb work on OpenBSD, not as simple as just
"configure" and "make".

Anyway, thanks very much again for your time and help!

Best Regards

Nan Xiao

On Thu, Jan 30, 2020 at 12:21 PM Simon Marchi <simark@simark.ca> wrote:
>
> On 2020-01-29 12:44 a.m., Nan Xiao wrote:
> > Hi gdb community,
> >
> > Greeting from me!
> >
> > I am building the newest gdb 8.3 on OpenBSD 6.6 server (Virtual
> > machine, amd64). The build is OK, but the gdb will crash when I tried
> > to use it to debug program:
> >
> > (gdb) bt
> > #0  0x0000000000000000 in ?? ()
> > #1  0x00000ec41fa1c506 in x86_dr_stopped_data_address
> > (state=0xec6da0e4010, addr_p=0x7f7ffffbde70) at
> > ../../gdb/nat/x86-dregs.c:608
> > #2  0x00000ec41fa1c7a3 in x86_dr_stopped_by_watchpoint
> > (state=0xec6da0e4010) at ../../gdb/nat/x86-dregs.c:650
> > #3  0x00000ec41f82ae47 in
> > x86_nat_target<obsd_nat_target>::stopped_by_watchpoint
> > (this=0xec6da0e4010) at ../../gdb/x86-nat.h:100
> > #4  0x00000ec41f84ff71 in watchpoints_triggered (ws=<optimized out>)
> > at ../../gdb/breakpoint.c:4727
> > #5  0x00000ec41f9ccb70 in handle_signal_stop (ecs=0x7f7ffffbe2f0) at
> > ../../gdb/infrun.c:5605
> > #6  0x00000ec41f9c1eeb in handle_inferior_event (ecs=0x7f7ffffbe2f0)
> > at ../../gdb/infrun.c:5178
> > #7  0x00000ec41f9c329b in fetch_inferior_event (client_data=<optimized
> > out>) at ../../gdb/infrun.c:3753
> > #8  0x00000ec41f95ce59 in gdb_do_one_event () at ../../gdb/event-loop.c:1064
> > #9  0x00000ec41f95d42d in start_event_loop () at ../../gdb/event-loop.c:371
> > #10 0x00000ec41f9f3392 in captured_command_loop () at ../../gdb/main.c:331
> > #11 0x00000ec41f9f2e13 in captured_main (data=<optimized out>) at
> > ../../gdb/main.c:1182
> > #12 gdb_main (args=<optimized out>) at ../../gdb/main.c:1198
> > #13 0x00000ec41f7fc378 in main (argc=<optimized out>,
> > argv=0x7f7ffffbde70) at ../../gdb/gdb.c:32
> > (gdb) frame 1
> > #1  0x00000ec41fa1c506 in x86_dr_stopped_data_address
> > (state=0xec6da0e4010, addr_p=0x7f7ffffbde70) at
> > ../../gdb/nat/x86-dregs.c:608
> > 608   status = x86_dr_low_get_status ();
> >
> >
> > I tried to step-in debug it, and found it seems to access an invalid
> > function address:
> >
> > (top-gdb) display /i $pc
> > 1: x/i $pc
> > => 0x76de4af4f5 <x86_dr_stopped_data_address(x86_debug_reg_state*,
> > unsigned long*)+37>:
> >     mov    0x394874(%rip),%r12        # 0x76de843d70
> > (top-gdb) ni
> > 0x00000076de4af4fc 608   status = x86_dr_low_get_status ();
> > 1: x/i $pc
> > => 0x76de4af4fc <x86_dr_stopped_data_address(x86_debug_reg_state*,
> > unsigned long*)+44>: mov    0x18(%r12),%r11
> > (top-gdb)
> > 0x00000076de4af501 608   status = x86_dr_low_get_status ();
> > 1: x/i $pc
> > => 0x76de4af501 <x86_dr_stopped_data_address(x86_debug_reg_state*,
> > unsigned long*)+49>: callq  0x76de28f150 <__llvm_retpoline_r11>
> > (top-gdb)
> >
> > Program received signal SIGSEGV, Segmentation fault.
> > 0x0000000000000000 in ?? ()
> > 1: x/i $pc
> > => 0x0: <error: Cannot access memory at address 0x0>
> >
> > I checked the source code, but can't figure out the reasons. Anyone
> > can give some clues? Thanks very much in advance!
> >
> > Best Regards
> > Nan Xiao
> >
>
> The line where you are stopped at is nat/x86-dregs.c:608, which is:
>
>   status = x86_dr_low_get_status ();
>
> x86_dr_low_get_status is actually a macro defined to:
>
>   #define x86_dr_low_get_status() (x86_dr_low.get_status ())
>
> So try to print "x86_dr_low.get_status", I expected that we will see 0 / NULL.
>
> This callback is probably supposed to get set in _initialize_x86_bsd_nat:
>
>   void
>   _initialize_x86_bsd_nat ()
>   {
>   #ifdef HAVE_PT_GETDBREGS
>     x86_dr_low.set_control = x86bsd_dr_set_control;
>     x86_dr_low.set_addr = x86bsd_dr_set_addr;
>     x86_dr_low.get_addr = x86bsd_dr_get_addr;
>     x86_dr_low.get_status = x86bsd_dr_get_status;
>     x86_dr_low.get_control = x86bsd_dr_get_control;
>     x86_set_debug_register_length (sizeof (void *));
>   #endif /* HAVE_PT_GETDBREGS */
>   }
>
> However, this assignment is guarded by HAVE_PT_GETDBREGS.  Could you look into
> your gdb/config.h, what is the value of HAVE_PT_GETDBREGS?  If it's indeed undef'ed,
> can you look into your config.log, to see how configure came to the conclusion that
> you don't have that feature?
>
> It would also be good to confirm that this function (_initialize_x86_bsd_nat) is called
> at startup like it's supposed to.  This can be done by adding a printf in it, if you are
> unable to use gdb on itself.
>
> Simon