[Bug gdb/34430] New: [gdb] value_from_mpz misplaces big-endian DW_AT_data_bit_offset values

https://sourceware.org/bugzilla/show_bug.cgi?id=34430

            Bug ID: 34430
           Summary: [gdb] value_from_mpz misplaces big-endian
                    DW_AT_data_bit_offset values
           Product: gdb
           Version: HEAD
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: gdb
          Assignee: unassigned at sourceware dot org
          Reporter: firmiana402 at gmail dot com
  Target Milestone: ---

In gdb/value.c, value_from_mpz writes bit-precise integers by shifting
the logical value by type->bit_offset () before storing it:

    if (type->bit_size_differs_p ())
      {
        unsigned bit_off = type->bit_offset ();
        unsigned bit_size = type->bit_size ();

        storage.mask (bit_size);
        storage <<= bit_off;
      }

Using the raw DW_AT_data_bit_offset as a low-bit shift is not correct
for big-endian storage.

For example, consider this base type:

    DW_TAG_base_type
      DW_AT_encoding         DW_ATE_unsigned
      DW_AT_byte_size        2
      DW_AT_bit_size         8
      DW_AT_data_bit_offset  8
      DW_AT_endianity        DW_END_big

DWARF v5 Section 5.1 defines DW_AT_data_bit_offset as an offset from
the beginning of the containing storage.  The logical value therefore
occupies the second byte of the two-byte big-endian storage.  In the
low-bit numbering used by the gdb_mpz storage value, the required
shift is:

    16 - 8 - 8 = 0

The current code instead shifts by 8.  Writing 0xbb therefore creates
the bytes [0xbb, 0x00] rather than [0x00, 0xbb].

The later gdb_mpz::truncate call, defined in gdb/gmp-utils.h, only
serializes the already-shifted integer in the requested byte order.
It has no bit-offset information and cannot correct this placement.

Before shifting, value_from_mpz should convert a big-endian data bit
offset to an offset from the least significant bit:

    if (type_byte_order (type) == BFD_ENDIAN_BIG)
      bit_off = (type->length () * TARGET_CHAR_BIT
                 - bit_off - bit_size);

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