[help] Linux kernel debug error: Cannot find DIE

[help] Linux kernel debug error: Cannot find DIE

[朱庆]

Hi all,

I met an error while debugging open64 built linux kernel with gdb,
following steps and error msg:
(gdb) file vmlinux
(gdb) target remote ip_addr:1234
Remote debugging using ip_addr:1234
Dwarf Error: Cannot find DIE at 0x1634a7 referenced from DIE at
0x17a459 [in module linux-kernel/linux-2.6.32.6/Build/vmlinux]
And there is also such error when set break points.

I'm not sure this is caused by incorrect dwarf info generated by
open64 compiler or gdb issues, since the gcc build kernel binary is
able to debug.
I compared the dwarf info output of the two compiler with objdump
--dwarf=info , open64's binary is about 1.95GB and gcc's binary is
1.72GB, is that too large?

Can someone give any clue on this to determine what is wrong here?

Thanks ahead!
zhuqing

Re: [help] Linux kernel debug error: Cannot find DIE

[Tom Tromey]

>>>>> ">" == 朱庆  <zqing1986@gmail.com> writes:

>> Dwarf Error: Cannot find DIE at 0x1634a7 referenced from DIE at
>> 0x17a459 [in module linux-kernel/linux-2.6.32.6/Build/vmlinux]

>> I'm not sure this is caused by incorrect dwarf info generated by
>> open64 compiler or gdb issues, since the gcc build kernel binary is
>> able to debug.
>> I compared the dwarf info output of the two compiler with objdump
>> --dwarf=info , open64's binary is about 1.95GB and gcc's binary is
>> 1.72GB, is that too large?

>> Can someone give any clue on this to determine what is wrong here?

There isn't an easy way to know what has gone wrong here.

It could be incorrect DWARF -- meaning a compiler or (less likely)
linker bug.  Or, it could be a gdb bug.

I would debug it by using objdump or readelf to look at the DWARF around
the DIEs mentioned in the error message.

The size of the DWARF is probably not an issue.
At least, this is not the error I would expect in that situation.

Tom

Re: [help] Linux kernel debug error: Cannot find DIE

[朱庆]

I accept your suggestion and have a look at dwarf dump, and find some
interestings:
I just write very 2 simple case, and link them.
There is warning like with the linked object:
objdump: Warning: Bogus end-of-siblings marker detected at offset 7a
in .debug_info section
Does this warning will affect debugging that causes errors I met?

Thanks
zhuqing

2012/7/12 Tom Tromey <tromey@redhat.com>:
>>>>>> ">" == 朱庆  <zqing1986@gmail.com> writes:
>
>>> Dwarf Error: Cannot find DIE at 0x1634a7 referenced from DIE at
>>> 0x17a459 [in module linux-kernel/linux-2.6.32.6/Build/vmlinux]
>
>>> I'm not sure this is caused by incorrect dwarf info generated by
>>> open64 compiler or gdb issues, since the gcc build kernel binary is
>>> able to debug.
>>> I compared the dwarf info output of the two compiler with objdump
>>> --dwarf=info , open64's binary is about 1.95GB and gcc's binary is
>>> 1.72GB, is that too large?
>
>>> Can someone give any clue on this to determine what is wrong here?
>
> There isn't an easy way to know what has gone wrong here.
>
> It could be incorrect DWARF -- meaning a compiler or (less likely)
> linker bug.  Or, it could be a gdb bug.
>
> I would debug it by using objdump or readelf to look at the DWARF around
> the DIEs mentioned in the error message.
>
> The size of the DWARF is probably not an issue.
> At least, this is not the error I would expect in that situation.
>
> Tom

Re: [help] Linux kernel debug error: Cannot find DIE

[Tom Tromey]

>>>>> ">" == 朱庆  <zqing1986@gmail.com> writes:

>> I accept your suggestion and have a look at dwarf dump, and find some
>> interestings:
>> I just write very 2 simple case, and link them.
>> There is warning like with the linked object:
>> objdump: Warning: Bogus end-of-siblings marker detected at offset 7a
>> in .debug_info section
>> Does this warning will affect debugging that causes errors I met?

It is hard to say for sure what the effect will be, but it does sound
like your compiler or linker has a bug.

Tom