Cannot insert breakpoint -1.

Cannot insert breakpoint -1.

[Anmol P. Paralkar]

Hello,

  Is there a reason why GDB will try to set a breakpoint (breakpoint number: -1) at
  the entry point address on a 'c', after a 'target remote'? (I am using a GDB stub
  at the remote end that resides in a hypervisor upon which guest OS'es run).

  I am using gdb-6.8.50.20090717; but, I suppose that this is general GDB behaviour.

--------------------------------------------------------------------------------
  (gdb) target remote :12007
Remote debugging using :12007
warning: Loadable segment "cpu0_heap_stack" outside of ELF segments
warning: Loadable segment "cpu1_heap_stack" outside of ELF segments
warning: Loadable segment "cpu2_heap_stack" outside of ELF segments
warning: Loadable segment "cpu3_heap_stack" outside of ELF segments
warning: Loadable segment "cpu4_heap_stack" outside of ELF segments
warning: Loadable segment "cpu5_heap_stack" outside of ELF segments
warning: Loadable segment "cpu6_heap_stack" outside of ELF segments
warning: Loadable segment ".stats_heap" outside of ELF segments
warning: Loadable segment ".dpa_mem" outside of ELF segments
0x0013c880 in ?? ()
(gdb) p $pc
$1 = (void (*)()) 0x13c880
(gdb) c
Continuing.
Warning:
Cannot insert breakpoint -1.
Error accessing memory address 0x110000: Unknown error 18446744073709551615.
--------------------------------------------------------------------------------

  (The entry point is likely not mapped into the TLB at that stage and so it is
  really a visible problem, else it just does this silently without it getting
  any attention).

  I tried to work around the issue by using set remote-<name>-packet, where <name>
  equalled software-breakpoint - but that does not work (it did not send a Z0, but
  sent an X packet instead (verified with an earlier GDB compiled with tracing on)):

--------------------------------------------------------------------------------
(gdb) c
Continuing.
^C
Program received signal SIGTRAP, Trace/breakpoint trap.
0x0013c888 in ?? ()
(gdb) c
Continuing.
Warning:
Cannot insert breakpoint -1.
Error accessing memory address 0x110000: Input/output error.
--------------------------------------------------------------------------------

  I then also re-tried with <name> equal to binary-download - but that does not work
  (it did not send the X, but sent an M packet instead):

--------------------------------------------------------------------------------
(gdb) c
Continuing.
^C
Program received signal SIGTRAP, Trace/breakpoint trap.
Cannot remove breakpoints because program is no longer writable.
Further execution is probably impossible.
0x0013c880 in ?? ()
(gdb) 
--------------------------------------------------------------------------------

  Please could someone kindly shed light on this GDB behaviour?

  Thank you.

Anmol.

Re: Cannot insert breakpoint -1.

[Daniel Jacobowitz]

On Fri, Jul 24, 2009 at 04:12:08PM -0500, Anmol P. Paralkar wrote:
> Hello,
> 
>  Is there a reason why GDB will try to set a breakpoint (breakpoint number: -1) at
>  the entry point address on a 'c', after a 'target remote'? (I am using a GDB stub
>  at the remote end that resides in a hypervisor upon which guest OS'es run).
> 
>  I am using gdb-6.8.50.20090717; but, I suppose that this is general GDB behaviour.

What target is it configured for?

"maint info break" will give you some more information.  This is
probably a shared library event breakpoint.
>  (The entry point is likely not mapped into the TLB at that stage and so it is
>  really a visible problem, else it just does this silently without it getting
>  any attention).

IMO it's a bug in your stub if you can't set breakpoints when not
mapped into the TLB...

-- 
Daniel Jacobowitz
CodeSourcery

Re: Cannot insert breakpoint -1.

[Anmol P. Paralkar]

On Fri, 24 Jul 2009, Daniel Jacobowitz wrote:

> On Fri, Jul 24, 2009 at 04:12:08PM -0500, Anmol P. Paralkar wrote:
>> Hello,
>>
>>  Is there a reason why GDB will try to set a breakpoint (breakpoint number: -1) at
>>  the entry point address on a 'c', after a 'target remote'? (I am using a GDB stub
>>  at the remote end that resides in a hypervisor upon which guest OS'es run).
>>
>>  I am using gdb-6.8.50.20090717; but, I suppose that this is general GDB behaviour.
>
> What target is it configured for?

  The target is: powerpc-linux-gnu

  The application is not a linux application, but a "guest" OS that runs on top of the
  hypervisor.

>
> "maint info break" will give you some more information.  This is
> probably a shared library event breakpoint.

  Yes, I see:

(gdb) maintenance info breakpoints 
Num     Type           Disp Enb Address    What
-1      shlib events   keep y   0x00110000 <_start>

  How do I delete it/prevent it from being set? Is this a configure time setting?

  I tried:

(gdb) delete -1
negative value

  --

  Thanks a lot!

Anmol.

Re: Cannot insert breakpoint -1.

[Daniel Jacobowitz]

On Fri, Jul 24, 2009 at 05:00:45PM -0500, Anmol P. Paralkar wrote:
>  Yes, I see:
> 
> (gdb) maintenance info breakpoints Num     Type           Disp Enb
> Address    What
> -1      shlib events   keep y   0x00110000 <_start>
> 
>  How do I delete it/prevent it from being set? Is this a configure time setting?

It may help to use a powerpc-elf debugger instead.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Cannot insert breakpoint -1.

[Anmol P. Paralkar]

On Fri, 24 Jul 2009, Daniel Jacobowitz wrote:

> On Fri, Jul 24, 2009 at 05:00:45PM -0500, Anmol P. Paralkar wrote:
>>  Yes, I see:
>>
>> (gdb) maintenance info breakpoints Num     Type           Disp Enb
>> Address    What
>> -1      shlib events   keep y   0x00110000 <_start>
>>
>>  How do I delete it/prevent it from being set? Is this a configure
> time setting?
>
> It may help to use a powerpc-elf debugger instead.

  I tried powerpc-elf GDB, and it'll work (I see that I cannot set a
  breakpoint in the application - but that seems to be a problem with
  the application itself).

  --

  I do have one question about powerpc-linux-gnu GDB; I see that the breakpoint
  is inserted inspite of:

(gdb) show stop-on-solib-events
Stopping for shared library events is 0.

  --

  From breakpoint.h:

     /* Some dynamic linkers (HP, maybe Solaris) can arrange for special
        code in the inferior to run when significant events occur in the
        dynamic linker (for example a library is loaded or unloaded).

        By placing a breakpoint in this magic code GDB will get control
        when these significant events occur.  GDB can then re-examine
        the dynamic linker's data structures to discover any newly loaded
        dynamic libraries.  */
     bp_shlib_event,

  --

  I would assume that if stop-on-solib-events is 0, then it ought not to set
  the -1 breakpoint. Please could you help understanding this part.

Thank you,
Anmol.

Re: Cannot insert breakpoint -1.

[Michael Snyder]

Anmol P. Paralkar wrote:
> On Fri, 24 Jul 2009, Daniel Jacobowitz wrote:
> 
>> On Fri, Jul 24, 2009 at 05:00:45PM -0500, Anmol P. Paralkar wrote:
>>>  Yes, I see:
>>>
>>> (gdb) maintenance info breakpoints Num     Type           Disp Enb
>>> Address    What
>>> -1      shlib events   keep y   0x00110000 <_start>
>>>
>>>  How do I delete it/prevent it from being set? Is this a configure
>> time setting?
>>
>> It may help to use a powerpc-elf debugger instead.
> 
>   I tried powerpc-elf GDB, and it'll work (I see that I cannot set a
>   breakpoint in the application - but that seems to be a problem with
>   the application itself).
> 
>   --
> 
>   I do have one question about powerpc-linux-gnu GDB; I see that the breakpoint
>   is inserted inspite of:
> 
> (gdb) show stop-on-solib-events
> Stopping for shared library events is 0.
> 
>   --
> 
>   From breakpoint.h:
> 
>      /* Some dynamic linkers (HP, maybe Solaris) can arrange for special
>         code in the inferior to run when significant events occur in the
>         dynamic linker (for example a library is loaded or unloaded).
> 
>         By placing a breakpoint in this magic code GDB will get control
>         when these significant events occur.  GDB can then re-examine
>         the dynamic linker's data structures to discover any newly loaded
>         dynamic libraries.  */
>      bp_shlib_event,
> 
>   --
> 
>   I would assume that if stop-on-solib-events is 0, then it ought not to set
>   the -1 breakpoint. Please could you help understanding this part.

Anmol,

"stop-on-solib-events" means 'stop' as in "show the user a gdb prompt".

If it's unset, we still insert the breakpoints so that we can
handle the shared library events (read symbol files), we just
don't go to the interactive prompt.

I think it's really unusual for someone to want to actually
go to the prompt when a shared lib event occurs, but normally
we want symbols to be loaded.  Don't know if there's any way
to actually disable that...

Michael