[llvm-dev] llvm-objdump: failed to parse debug information

Jerome Forissier via llvm-dev llvm-dev at lists.llvm.org
Tue Apr 28 11:14:30 PDT 2020



On 4/28/20 6:19 PM, Peter Smith wrote:
>> Hi,
>>
>> In a 32-bit ARM build, I am seeing the following warning (edited for
>> simplicity, I can provide full logs if necessary):
>>
>>> llvm-objdump -l -d -x file.elf
>>> llvm-objdump: warning: 'file.elf': failed to parse debug information for file.elf
>>
>>
>> All object files and static libraries seem to have debug info (i.e.,
>> llvm-objdump does not complain when run on each file individually and
>> the disassembly output shows file/line information).
>>
>> In order to identify where the ELF file is lacking debug info, I added
>> some debug traces to llvm-objdump and it seems the following symbol is
>> the culprit: __ThumbV7PILongThunk_<my_func>
>>
>> Is this expected behavior? How to fix it?
> 
> It is expected that __ThumbV7PILongThunk_<my_func> has no debug information. A thunk is a linker generated code-sequence to extend the range of a branch. Usually we would want a debugger to step over these bits of code as they are largely there to bridge between functions the user wrote and are not interesting in their own right. There is also the practical consideration that there is no source code for these code-sequences; and as they are called between procedures they are limited in what state they can modify. For example a Thunk in Arm is only permitted to change the inter-procedural scratch register r12.
> 
> I would be surprised if missing debug information for a thunk would cause llvm-objdump to give an error message.

The reason I mentioned the thunk is because I added debug traces like so:

====
diff --git a/llvm/tools/llvm-objdump/llvm-objdump.cpp b/llvm/tools/llvm-objdump/llvm-objdump.cpp
index 92c130be113..ae1c02d49e8 100644
--- a/llvm/tools/llvm-objdump/llvm-objdump.cpp
+++ b/llvm/tools/llvm-objdump/llvm-objdump.cpp
@@ -638,7 +638,9 @@ void SourcePrinter::printSourceLine(raw_ostream &OS,
   if (LineInfo.FileName == DILineInfo::BadString) {
     if (!WarnedNoDebugInfo) {
       std::string Warning =
-          "failed to parse debug information for " + ObjectFilename.str();
+          "failed to parse debug information for " + ObjectFilename.str() +
+	  " Address " + to_string(Address.Address) + " SectionIndex " +
+	  to_string(Address.SectionIndex);
       if (!ErrorMessage.empty())
         Warning += ": " + ErrorMessage;
       reportWarning(Warning, ObjectFilename);
====

Which gives:

$ llvm-objdump -l ldelf.elf >/dev/null
llvm-objdump: warning: 'ldelf.elf': failed to parse debug information for ldelf.elf Address 19404 SectionIndex 1

And readelf reports (note that 0x4bcc == 19404):

$ arm-linux-gnueabihf-readelf -a ldelf.elf | grep -E '(] \.text|<.*Thunk)'
  [ 1] .text             PROGBITS        00000000 001000 004bd8 00  AX  0   0  4
0x4bcc <__ThumbV7PILongThunk_utee_log>: 0x1 [cantunwind]

-- 
Jerome

> 
> 
>> Thanks,
>> --
>> Jerome
> 
> ________________________________________
> From: llvm-dev <llvm-dev-bounces at lists.llvm.org> on behalf of Jerome Forissier via llvm-dev <llvm-dev at lists.llvm.org>
> Sent: 28 April 2020 15:23
> To: llvm-dev at lists.llvm.org
> Subject: [llvm-dev] llvm-objdump: failed to parse debug information
> 
> Hi,
> 
> In a 32-bit ARM build, I am seeing the following warning (edited for
> simplicity, I can provide full logs if necessary):
> 
>> llvm-objdump -l -d -x file.elf
>> llvm-objdump: warning: 'file.elf': failed to parse debug information for file.elf
> 
> 
> All object files and static libraries seem to have debug info (i.e.,
> llvm-objdump does not complain when run on each file individually and
> the disassembly output shows file/line information).
> 
> In order to identify where the ELF file is lacking debug info, I added
> some debug traces to llvm-objdump and it seems the following symbol is
> the culprit: __ThumbV7PILongThunk_<my_func>
> 
> Is this expected behavior? How to fix it?
> 
> LLVM version is llvmorg-11-init-12683-g54b3f91d205 but AFAICT anything
> above 10.0 behaves the same.
> 
> Thanks,
> --
> Jerome
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
> 


More information about the llvm-dev mailing list