Reverse debugging for arm baremetal targets?

Reverse debugging for arm baremetal targets?

[Terry Guo]

Hi there,

Is this feature enabled for arm baremetal targets? If not, is it
possible to do so and what kind of problems we need to solve to
achieve this? Thanks in advance for your help.

Best Regards,
Terry

Re: Reverse debugging for arm baremetal targets?

[Hui Zhu]

On Mon, Jul 22, 2013 at 9:37 AM, Terry Guo <flameroc@gmail.com> wrote:
> Hi there,
>
> Is this feature enabled for arm baremetal targets? If not, is it
> possible to do so and what kind of problems we need to solve to
> achieve this? Thanks in advance for your help.
>
> Best Regards,
> Terry

If just for reverse debug, I think it is not depend on the arch.  So I
think it is OK for ARM.

And if I remember is right, simics support it.

Thanks,
Hui

Re: Reverse debugging for arm baremetal targets?

[Terry Guo]

On Mon, Jul 22, 2013 at 10:07 AM, Hui Zhu <teawater@gmail.com> wrote:
> On Mon, Jul 22, 2013 at 9:37 AM, Terry Guo <flameroc@gmail.com> wrote:
>> Hi there,
>>
>> Is this feature enabled for arm baremetal targets? If not, is it
>> possible to do so and what kind of problems we need to solve to
>> achieve this? Thanks in advance for your help.
>>
>> Best Regards,
>> Terry
>
> If just for reverse debug, I think it is not depend on the arch.  So I
> think it is OK for ARM.
>
> And if I remember is right, simics support it.
>
> Thanks,
> Hui

Thanks for your reply. I did some tests and it seems to me that
reverse debugging doesn't work for baremetal targets. My test includes
a ARM Cortex-M3 board running a baremetal program, a gdb server and a
baremetal gdb. The gdb version is:

GNU gdb (GDB) 7.6

After I connect to M3 board and run command "record", I got:

Breakpoint 1, main () at main.cpp:24
24 xyz = 0;
(gdb) record
Process record: the current architecture doesn't support record function.
(gdb) show ar
architecture  args          arm
(gdb) show architecture
The target architecture is set automatically (currently arm)
(gdb) n
26 foo ();
(gdb) s
foo () at main.cpp:16
16  xyz = 1;
(gdb) n
17  return bar ();
(gdb) s
bar () at main.cpp:10
10  xyz = 2;
(gdb) reverse-next
Target remote does not support this command.

Am I doing something wrong? Please advise.

BR,
Terry

Re: Reverse debugging for arm baremetal targets?

[Jan Kratochvil]

On Mon, 22 Jul 2013 08:10:33 +0200, Terry Guo wrote:
> Thanks for your reply. I did some tests and it seems to me that
> reverse debugging doesn't work for baremetal targets.

It looks so:

$ grep '  set_gdbarch_process_record ' gdb/*
amd64-linux-tdep.c:  set_gdbarch_process_record (gdbarch, i386_process_record);
arm-linux-tdep.c:  set_gdbarch_process_record (gdbarch, arm_process_record);
i386-linux-tdep.c:  set_gdbarch_process_record (gdbarch, i386_process_record);
moxie-tdep.c:  set_gdbarch_process_record (gdbarch, moxie_process_record);

But the OS dependent part seems to be missing there:
arm-tdep.h:
   /* Parse swi insn args, sycall record.  */
  int (*arm_swi_record) (struct regcache *regcache);
 - which does not seem to be set anywhere
 
So the current set_gdbarch_process_record initialization could be possibly
moved to arm-tdep.c.  But I did not play more with it.


Jan

Re: Reverse debugging for arm baremetal targets?

[Hui Zhu]

On Mon, Jul 22, 2013 at 2:10 PM, Terry Guo <flameroc@gmail.com> wrote:
> On Mon, Jul 22, 2013 at 10:07 AM, Hui Zhu <teawater@gmail.com> wrote:
>> On Mon, Jul 22, 2013 at 9:37 AM, Terry Guo <flameroc@gmail.com> wrote:
>>> Hi there,
>>>
>>> Is this feature enabled for arm baremetal targets? If not, is it
>>> possible to do so and what kind of problems we need to solve to
>>> achieve this? Thanks in advance for your help.
>>>
>>> Best Regards,
>>> Terry
>>
>> If just for reverse debug, I think it is not depend on the arch.  So I
>> think it is OK for ARM.
>>
>> And if I remember is right, simics support it.
>>
>> Thanks,
>> Hui
>
> Thanks for your reply. I did some tests and it seems to me that
> reverse debugging doesn't work for baremetal targets. My test includes
> a ARM Cortex-M3 board running a baremetal program, a gdb server and a
> baremetal gdb. The gdb version is:
>
> GNU gdb (GDB) 7.6
>
> After I connect to M3 board and run command "record", I got:

This error is about process record but not reverse debug.

>
> Breakpoint 1, main () at main.cpp:24
> 24 xyz = 0;
> (gdb) record
> Process record: the current architecture doesn't support record function.

This is the reason why you cannot use reverse debug commands.

You got this error is because:
  /* Reversible debugging, process record.  */
  set_gdbarch_process_record (gdbarch, arm_process_record);
This line is called in arm_linux_init_abi but you are not a linux target.

I found all the "set_gdbarch_process_record" is called in
xxx_linux_init function make it depend on Linux.
I forget why I do it.  :P
I think 2 reasons let me do it:
1. syscall record just support linux.
2. make it can pass review.  :P

But most of prec function is just to record the behavior of
instruction.  So maybe we can move it to arch init code.

Thanks,
Hui

> (gdb) show ar
> architecture  args          arm
> (gdb) show architecture
> The target architecture is set automatically (currently arm)
> (gdb) n
> 26 foo ();
> (gdb) s
> foo () at main.cpp:16
> 16  xyz = 1;
> (gdb) n
> 17  return bar ();
> (gdb) s
> bar () at main.cpp:10
> 10  xyz = 2;
> (gdb) reverse-next
> Target remote does not support this command.
>
> Am I doing something wrong? Please advise.




>
> BR,
> Terry

Re: Reverse debugging for arm baremetal targets?

[Yao Qi]

On 07/22/2013 02:38 PM, Jan Kratochvil wrote:
> But the OS dependent part seems to be missing there:
> arm-tdep.h:
>     /* Parse swi insn args, sycall record.  */
>    int (*arm_swi_record) (struct regcache *regcache);
>   - which does not seem to be set anywhere
>

I raised this question during the code review, and Oza (the author) 
wanted to do them in phase 3, which handles OS related stuff, such as 
syscall.  What we have in trunk is phase 2.

> So the current set_gdbarch_process_record initialization could be possibly
> moved to arm-tdep.c.  But I did not play more with it.

Right, the existing Oza's work in trunk is about ARM reversed debugging, 
without OS stuff.  Probably we should call set_gdbarch_process_record in 
arm-tdep.c, but not sure how good or bad the results are.

-- 
Yao (齐尧)

Re: Reverse debugging for arm baremetal targets?

[Hui Zhu]

On Mon, Jul 22, 2013 at 3:29 PM, Yao Qi <yao@codesourcery.com> wrote:
> On 07/22/2013 02:38 PM, Jan Kratochvil wrote:
>>
>> But the OS dependent part seems to be missing there:
>> arm-tdep.h:
>>     /* Parse swi insn args, sycall record.  */
>>    int (*arm_swi_record) (struct regcache *regcache);
>>   - which does not seem to be set anywhere
>>
>
> I raised this question during the code review, and Oza (the author) wanted
> to do them in phase 3, which handles OS related stuff, such as syscall.
> What we have in trunk is phase 2.
>
>
>> So the current set_gdbarch_process_record initialization could be possibly
>> moved to arm-tdep.c.  But I did not play more with it.
>
>
> Right, the existing Oza's work in trunk is about ARM reversed debugging,
> without OS stuff.  Probably we should call set_gdbarch_process_record in
> arm-tdep.c, but not sure how good or bad the results are.

I think you should call it arm process record but not reverse debug.

Reverse debug depend on the target support but I don't think call prec
as reverse debug is right.

Thanks,
Hui

>
> --
> Yao (齐尧)