This is the mail archive of the gdb@sources.redhat.com mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

gas (or maybe gdb/gcc) problem with o64 Mips and DWARF


I don't understand how to fix this, but here's what I'm seeing.

Configuring and building for mips64-elf.  The default abi is o64 which
has 32-bit pointers. I cannot debug anything because GDB cannot
correctly read the linetable.  readelf can read it but dwarfdump cannot.
Looking more carefully, I see that 
DW_LNE_set_address has a 64-bit address, but the compilation units all
set the address size to 32 bits.  This causes a problem for GDB's dwarf
reader.

To reproduce the problem, compile any small program with mips64-elf-gcc
then attempt to debug with gdb or insight.  Line numbers will either be
nonexistent or incorrect.

The problem appears to be in gas/dwarf2dbg.c in function dwarf2_finish()
which has the following:

 /* Calculate the size of an address for the target machine.  */
  sizeof_address = bfd_arch_bits_per_address (stdoutput) / 8;

So gas is not getting address size from the compilation unit header, but
from the architecture, which for mips64-elf is "mips:4000".  So how
should this work?  I cannot see how to grab info from the CU header. And
I don't know how else to find out from GCC what size a pointer is.

(What happens if I add "-mlong64" to the compile line?  Now gcc
generates CUs with 64-bit addresses and gas agrees and all works well.)

Martin







Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]