Hi Xavier,
thanks for the info. The architecture is has IA32 instruction set and the compiler is proprietary
(sorry, some confidentiality here). From TCF:s perspective, the architecture is identified
as known
and has a stack crawler.
>If I understand your issue, it seems that the stack walker is not able to compute the regs down for your frame.
>I would start by doing a readelf -wf or readelf -wF.
>This is the dump of rules in .debug_frames.
Yes, that is the problem. I have not studied the .debug_frame yet, but will do this.
>If you're working with an existing architecture, implied rules might be incomplete.
This looks plausible. Perhaps the reason this has not been discovered before is that mainly GCC code
has been debugged, and not the variant of DWARF information produced by this compiler.
Again, thank you, now I have some topics to investigate.
Regards,
Claes
From: tcf-dev-bounces@xxxxxxxxxxx [mailto:tcf-dev-bounces@xxxxxxxxxxx]
On Behalf Of Xavier Pouyollon
Sent: Tuesday, February 9, 2016 8:57 AM
To: TCF Development <tcf-dev@xxxxxxxxxxx>
Subject: Re: [tcf-dev] No registers setup in Frame setup from DWARF info
Hi Claes,
Which architecture are you working with ? Which compiler ?
> If I change the code so that stack crawler (crawl_stack_frame()) is always used to setup ‘down´, I get a full stack.
To get a backtrace, you have:
- the stack walker that uses the .debug_frame information
- the stack crawler which does NOT use .debug_frame information but tries to guess how registers are saved by going down
to the epilogue (for instance /agent/machine/arm/tcf/stack-crawl-arm.c).
If I understand your issue, it seems that the stack walker is not able to compute the regs down for your frame.
I would start by doing a readelf -wf or readelf -wF.
This is the dump of rules in .debug_frames.
Look at your current PC (instruction pointer). Do you see any rules for this address ?
Are these rules valid ? They may rely on an implict dwarf rules, not described in the CIE
(Common information entry). You can take a look in dwarfframe.c / get_reg.
Here, you can see some architecture specific implied rules.
Are you adding a new architecture ? If so, you may miss some implied rules.
If you're working with an existing architecture, implied rules might be incomplete.
You may miss some description in the generated CIE.
Either fix the CIE generation or add implied rules in dwarfframe.c
> reg->rule for return address register is always zero.
You may miss a implied rule here. Take a look at get_reg, case EM_ARM.
You'll see the default rule for return_address_register there
( if (n == rules.return_address_register) {
...)
You may also take a look here (some notes I took when working on CFA / Stack / Runcontrol stuff):
https://dev.eclipse.org/mhonarc/lists/tcf-dev/pdfDWqjnobpcm.pdf
First, try to compute the backtrace "by hand" using the dump of readelf -wF.
That will tell you if the generated rules are correct or not.
Hope it helps a little,
Best Regards,
Xavier.
On 08/02/2016 19:42, Lillieskold, Claes wrote:
Hi,
I’m having problem generating a stack frame longer than one.
The problem I’m seeing is in stack_trace.c:trace_stack() where get_next_stack_frame() returns success,
but doesn’t set down.has_reg_data, which leads to only one frame being produced.
If I change the code so that stack crawler (crawl_stack_frame()) is always used to setup ‘down´, I get a full stack.
I’m not sure why this happens and have been debugging symbols.elf:get_stack_tracing_info() without
understanding why the number of registers (info->reg_cnt) is zero for the frame.
No exceptions are thrown during the call to get_stack_tracing_info().
It seems that the number of registers in the frame are copied from dwarf_stack_trace_regs_cnt,
but this is always zero for me, and when trying to identify where dwarf_stack_trace_regs_cnt could
be set to non-zero, I find the function dwarfframe.c:generate_command(), where calling
functions that can increase dwarf_stack_trace_regs_cnt is never called because reg->rule for
return address register is always zero.
reg = get_reg(&frame_regs, rules.return_address_register);
if (reg->rule != 0) {
Sorry about the imprecise description, but any pointers or suggestions are very welcome.
Thanks,
Claes
----------------------------------------------------------------------
Intel Sweden AB
Registered Office: Isafjordsgatan 30B, 164 40 Kista, Stockholm, Sweden
Registration Number: 556189-6027
This e-mail and any attachments may contain confidential material for
the sole use of the intended recipient(s). Any review or distribution
by others is strictly prohibited. If you are not the intended
recipient, please contact the sender and delete all copies.
_______________________________________________
tcf-dev mailing list
tcf-dev@xxxxxxxxxxx
To change your delivery options, retrieve your password, or unsubscribe from this list, visit
https://dev.eclipse.org/mailman/listinfo/tcf-dev
