[Bug gdb/34363] New: [gdb] gdb ignores the DW_OP_regval_type base type operand inside DW_OP_entry_value

https://sourceware.org/bugzilla/show_bug.cgi?id=34363

            Bug ID: 34363
           Summary: [gdb] gdb ignores the DW_OP_regval_type base type
                    operand inside DW_OP_entry_value
           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: ---

DW_OP_entry_value matches its nested register block with
dwarf_block_to_dwarf_reg in gdb/dwarf2/expr.c, 
whose contract is documented as:

    /* If BLOCK contains DW_FORM_block* with single DW_OP_reg* return the
       DWARF register number.  Otherwise return -1.  */

In practice the function accepts more than a single DW_OP_reg*: it also matches
DW_OP_regval_type / DW_OP_GNU_regval_type (and DW_OP_regx). Accepting the typed
register form is fine per DWARF, but DW_OP_regval_type is handled incompletely.
Its branch reads the register operand and then merely skips the type operand
with gdb_skip_leb128, and the function returns only the register number:

    if (*buf == DW_OP_regval_type || *buf == DW_OP_GNU_regval_type)
      {
        buf++;
        buf = gdb_read_uleb128 (buf, buf_end, &dwarf_reg);
        if (buf == NULL)
          return -1;
        buf = gdb_skip_leb128 (buf, buf_end);
        if (buf == NULL)
          return -1;
      }
    ...
    return dwarf_reg;

The caller then uses the result as a plain register value
(CALL_SITE_PARAMETER_DWARF_REG). In DWARF the second operand of
DW_OP_regval_type is the base type the register contents are to be interpreted
as; dropping it means a typed, possibly narrower entry value is materialized as
the untyped full register value, silently changing the meaning of an expression
GDB has otherwise accepted.

This leaves DW_OP_regval_type half-supported: accepted, but stripped of the
operand that defines it. It should be handled one way or the other — either
reject the DW_OP_regval_type / DW_OP_GNU_regval_type form (consistent with the
"single DW_OP_reg*" contract the comment already states), or support it fully
by
applying the base type to the register value.

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