[Bug libdw/28220] New: dwarf_location_attr returns high-bit junk from .debug_addr when fetching 32-bit addresses

["gprocida at google dot com" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=28220

            Bug ID: 28220
           Summary: dwarf_location_attr returns high-bit junk from
                    .debug_addr when fetching 32-bit addresses
           Product: elfutils
           Version: unspecified
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: libdw
          Assignee: unassigned at sourceware dot org
          Reporter: gprocida at google dot com
                CC: elfutils-devel at sourceware dot org
  Target Milestone: ---

In libdw, there are snippets of code of the form:

dwarf_formaddr.c:
  71   if (address_size == 4)
  72     *addr = read_4ubyte_unaligned (dbg, datap);
  73   else
  74     *addr = read_8ubyte_unaligned (dbg, datap);

However, when I use the public interfaces in order to read locations, it goes
wrong when for Clang-built 32-bit objects with DWARF 5 debug information, with
extra junk in the upper 32 bits.

The attachment contains a full example. It's possible that I'm using the API
incorrectly. This is the debug output for the problematic case:

clang 5 32
bias=0
bias=0
dwarf=0x55f103cf8a10 elf=0x55f103cf5610
 version=5 unit_type=1 cu_die.addr=0x7fc3032e0078 cu_die.cu=0x55f103cf8e98
cu_die.abbrev=0 sub_die.addr=0 sub_die.cu=0 sub_die.abbrev=0
 unit_id=0 address_size=4 offset_size=4
  dwarf_string_attr(die, 1b)
 comp_dir=/usr/local/google/home/gprocida/dev/libabigail/b195152239
   child.addr=0x7fc3032e008a child.cu=0x55f103cf8e98 child.abbrev=0
    dwarf_string_attr(die, 3)
    dwarf_bool_attr(die, 3f)
    dwarf_location_attr(die, 2)
    attribute code=0x2 form=0x18 (expected=18)
     expr=0x55f103cf8ff8 exprlen=1
     op.atom=a1 op.number=0 op.number2=0 op.offset=0
     result code=0x11 form=0x1 (expected=1)
     address=0x1000400010000
   variable name=x name2=x external=1 location=1000400010000
   child.addr=0x7fc3032e0095 child.cu=0x55f103cf8e98 child.abbrev=0
   child.addr=0x7fc3032e0099 child.cu=0x55f103cf8e98 child.abbrev=0
    dwarf_string_attr(die, 3)
    dwarf_bool_attr(die, 3f)
    dwarf_location_attr(die, 2)
    attribute code=0x2 form=0x18 (expected=18)
     expr=0x55f103cf9048 exprlen=1
     op.atom=a1 op.number=1 op.number2=0 op.offset=0
     result code=0x11 form=0x1 (expected=1)
     address=0x6265440000010004
   variable name=y name2=y external=1 location=6265440000010004

Note that variable x's address includes variable y's address as the upper 32
bits and that variable y has junk in the upper 32 bits.

-- 
You are receiving this mail because:
You are on the CC list for the bug.