Issue 230524.1: Location Descriptions on the DWARF Stack

Author: Tony Tye, Cary Coutant
Champion: Cary Coutant
Date submitted: 2023-05-24
Date revised: 2025-04-11
Date closed: 2025-10-13
Type: Enhancement
Status: Accepted
DWARF version: 6

This is the 2025-04-11 version of the proposal. [ Compare with previous version ] [ Return to the latest version ]

This is one of several proposals for GPU support, extracted from AMD's work on DWARF Extensions for Heterogeneous Debugging for the LLVM compiler.

This is a large proposal, in part due to some document reorganization and terminology changes that permeate the DWARF spec. To help the reader see the proposed changes in context, a redline comparison of the affected parts of the DWARF spec is available.

Background

The DWARF 5 concept of location descriptions (Section 2.6) limits their use to cases where the location described is final, and not subject to some further modification, with two exceptions. First, if the location description is a memory location description, it is a simple DWARF expression (Section 2.5) that can be modified by further DWARF expression operators. Second, for any form of location description, it can be offset by a fixed number of bits by using a DW_OP_bit_piece composition operator.

Where do these limitations matter?

Consider the case of a FORTRAN array (as shown in Appendix D in D.2.1) that has been partially promoted to a register or registers. The evaluation of its lower and upper bounds depends on the location of the array as provided by DW_OP_push_object_address. If the array is not entirely in memory, this operation is not able to provide the address of the object, as it can only provide a memory address. If DW_OP_push_object_address were allowed to push a composite location description on the stack, we could apply further operations to locate the bounds of the array.

Similarly, consider the case of a pointer-to-member type in C++, where the object (or part of the object) has been promoted to a register. In this case, DW_AT_use_location is not able to provide the address of the object. In optimized code, it sometimes would need to provide a register location description or a composite location description, but these cannot be pushed onto the DWARF stack. If DW_AT_use_location were allowed to push a composite location description on the stack, we could apply further operations to determine the register location of the member being referenced.

Also consider the case where a DW_OP_call* operator is used to get the location of a variable. If the variable happens to be in a register at the current PC, the call operator cannot succeed, as it cannot push anything but a memory address on the stack.

All of these cases have a common limiting factor: that location descriptions cannot be pushed onto the stack, and subsequently operated on to produce derived location descriptions.

Overview

This proposal removes that limitation. A DWARF expression may evaluate to either a value or a location. Location descriptions are now simply DWARF expressions that evaluate to a location, and are now called location expressions.

The DWARF stack is extended so that it can hold elements that are either (typed) values or (single) locations. The operators in Section 2.6 that previously defined register and implicit locations are now considered part of a DWARF expression, and are no longer "terminal" in the sense that they cannot be part of a larger expression.

The literal encoding operations, defined in Section 2.5.1.1, push values onto the stack, except for DW_OP_addr and DW_OP_addrx, which push memory locations. These latter two operations are moved to a new section.

Stack operations, defined in Section 2.5.1.3, can operate on values or locations, or any combination of the two.

Most existing arithmetic and logical operators, defined in Section 2.5.1.4, continue to be limited to operating on values only.

The DW_OP_deref* operator is extended to operate on any location, and provide the value contained at that location, whether in memory, in a register, in implicit storage, or a composite value.

The DW_OP_push_object_address, renamed DW_OP_push_object_location operator pushes a location, which may be a memory address (as before), or a register, implicit storage, or a composite.

The DW_AT_use_location attribute provides an expression used to compute the address of a member for a pointer-to-member type, and expects the evaluation mechanism to provide the value of the pointer and the location of the object as implicitly-pushed elements on the stack. The latter element is now allowed to be any location.

Two new operators, DW_OP_offset and DW_OP_bit_offset, are introduced that allow a location on the stack to be modified by a byte or a bit offset.

The composite location operators, DW_OP_piece and DW_OP_bit_piece, are redefined to build up a composite location, which is held in the top element of the stack. A new operator, DW_OP_composite, is added to begin a new (empty) composite location.

The DW_OP_call* operators are now allowed to leave a location on the stack.

A new Section 2.5 "Values and Locations" is added, and the old Sections 2.5, "DWARF Expressions," and 2.6, "Location Descriptions," are moved into a new Chapter 3, "DWARF Expressions," and reorganized as follows:

Proposed Changes

Section 2.2 Attribute Types

In Table 2.3, Classes of attribute value, in the rows for "exprval" and "locdesc," replace the reference to Sections 2.5 and 2.6 with references to Chapter 3. In the row for "locdesc," change "locdesc" to "locexpr" and change "location description" to "location expression" (2 places).

Rename class locdesc to locexpr throughout the document.

Section (OLD) 2.5 DWARF Expressions [REMOVED]

This section is moved from Chapter 2 into a new Chapter 3.

Section (OLD) 2.6 Location Descriptions [REMOVED]

This section is moved from Chapter 2 into a new Chapter 3.

Section (NEW) 2.5 Values and Locations [NEW]

Add the following as a new subsection:

A DWARF expression is evaluated in a context that determines whether its result is expected to be a value or a location descriptor. Expressions that are expected to produce a location descriptor are called “location expressions.”

(In DWARF 5, the term “location description” was used to refer both to a sequence of DWARF operations that evaluates to a location, and to the result of that evaluation. For DWARF 6, we use the terms “location expression” for the expression and “location descriptor” for the result of evaluating the expression. A location descriptor describes the location of an object.)

Values on the stack are typed, and can represent a value of any supported base type of the target machine, or of the generic type, which is an integral type that has the size of an address in the default address space on the target machine, and unspecified signedness.

Debugging information must provide consumers a way to find the location of program variables, determine the bounds of dynamic arrays and strings, and possibly to find the base address of a subroutine’s stack frame or the return address of a subroutine. Furthermore, to meet the needs of recent computer architectures and optimization techniques, debugging information must be able to describe the location of an object whose location changes over the object’s lifetime.

Information about the location of program objects is provided by location expressions and location lists.

A location expression is a DWARF expression yielding a single location descriptor. These are sufficient for describing the location of any object as long as its lifetime is either static or the same as the lexical block that owns it, excluding any prologue or epilogue ranges, and it does not move during its lifetime. As the value of an attribute, a location expression is encoded using class locexpr.

A location list describes objects that have a limited lifetime or change their location during their lifetime. A location list is a list of location expressions, each associated with a range of program counters. Location lists are described in Section 3.18. As the value of an attribute, a location list is encoded using class loclist (which serves as an index into a separate section containing location lists).

A location list may have overlapping PC ranges, and thus may yield more than one location descriptor. In these cases, the object value stored in each location is expected to be the same (except for uninitialized/undefined parts of the value).

A location list that yields multiple locations can be used to describe objects that reside in more than one piece of storage at the same time. An object may have more than one location as a result of optimization. For example, a value that is only read may be promoted from memory to a register for some region of code, but later code may revert to reading the value from memory as the register may be used for other purposes.

When given multiple locations, a consumer can read the object’s value from any of those locations (since they all refer to storage that should contain the same value), but must write any changed value to all the locations.

DWARF can describe the location of program objects in several kinds of storage, such as a memory address space or a register. Each kind of storage is treated as a linear stream of bits of finite size, organized into bytes and/or words. The ordering of bits uses the bit numbering and direction conventions that are appropriate to the target architecture and the ABI. For example, on a little-endian architecture, bits are numbered within bytes and words from least-significant to most-significant (right-to-left), while on a big-endian architecture, bits are numbered left-to-right.

A location descriptor names a block of storage and a (non-negative, zero-based) bit offset relative to the start of that storage. It gives the location of the program object, which occupies a sequence of contiguous bits starting at that location. A location descriptor does not provide the size of the object, which is typically obtained from its type attribute.

DWARF can describe locations in five kinds of storage:

  • Memory storage. Corresponds to the target architecture memory address spaces. The location descriptor names an address space and provides an address within the address space as a bit offset. The offset must be less than the size (in bits) of the address space. There is always a default address space, and the target architecture may define additional address spaces. (The offset can be thought of as a byte or word address combined with a bit offset within the byte or word.)

  • Register storage. Corresponds to the target architecture registers. The location descriptor names a register and provides a bit offset within the register. The offset must be less than the size (in bits) of the register.

  • Undefined storage. Indicates no value is available, as when a variable has been optimized out. The location descriptor names an imaginary block of storage that cannot be read or written. The offset within this block of storage must be less than the size of the largest address space, but has no effect.

  • Implicit storage. Corresponds to fixed values that can only be read, as when a variable has been optimized out but can nevertheless be rematerialized from program context. The location descriptor names a block of imaginary, or virtual, storage and provides a bit offset within the block. The block is the same size as the fixed value that it holds, and the offset must be less than the size (in bits) of the block.

    A special form of implicit storage, created by the DW_OP_implicit_pointer operator (see Section 3.10), holds a virtual reference to a secondary location descriptor. Its size is the size of a pointer in the default memory address space, but no physical pointer is available. Reading (i.e., dereferencing) the location produces the secondary location descriptor. Because the contents of the storage cannot be represented as a sequence of bits, an offset other than 0 has no meaning and the location cannot be read.

  • Composite storage. DWARF can also describe composite storage, in which a program object is mapped to a mixture of locations in different kinds of storage, as when a field of a structure or a slice of an array is promoted to a register while the rest remains in memory. A composite location descriptor is a set of tuples of the form (S, E, L), where each tuple represents a mapping from bits in the half-open range [S, E) of the program object onto a location descriptor L. The offset D in a composite location descriptor refers to a bit position within the program object, selects a mapping where S ≤ D < E, and translates to the location L with an additional offset of D - S applied. In a composite location descriptor, each component tuple represents a disjoint range of bits (i.e., the ranges may not overlap). The offset must be less than the largest value of E among the tuples. In the process of fetching the value from a composite location, the consumer may need to reassemble bits from more than one of the mappings.

An implicit conversion between a memory location and a value may happen during the execution of any operation or when evaluation of the expression is completed. If a location is expected, but the result is a value, the value is implicitly treated as a memory address in the default address space, and converted to a memory location. If a value is expected, and the result is an addressable memory location (i.e., if the offset is a multiple of the byte or word size) in the default address space, the address is implicitly converted to a value of the generic type.

Chapter 3 DWARF Expressions [NEW]

Replace the contents of the preamble to the old Section 2.5 with:

DWARF expressions describe how to compute a value or specify a location. They are expressed in terms of DWARF operations that operate on a stack of elements, each of which may be either a value or a location.

A DWARF expression is encoded as a stream of operations, each consisting of an opcode followed by zero or more literal operands. The number of operands is implied by the opcode.

The result of a DWARF expression is the value or location on the top of the stack after evaluating the operations.

Values on the stack are typed, and can represent a value of any supported base type of the target machine, or of the generic type, which is an integral type that has the size of an address in the default address space on the target machine, and unspecified signedness.

Section 3.1 DWARF Expression Evaluation Context [was 2.5.1]

Move the contents of the old Section 2.5.1 DWARF Expression Evaluation Context here.

In the first paragraph, remove "and location descriptions (see Section ...)".

In item 1, "Required result kind," 2nd paragraph, change "location description" to "location".

In item 8, "Current program counter (PC)," 5th paragraph, change "only default location descriptions may be used" to "only default value or location list entries may be used."

In item 9, "Current object," 2nd paragraph, change "and by some attributes" to "and is implicitly defined by some attributes."

In the final (non-normative) paragraph of the section, change "A DWARF expression for a location description..." to "A DWARF expression...."

Section 3.2 Stack Operations [was 2.5.2.3 Stack Operations]

Move the contents of 2.5.2.3 Stack Operations here. Change the first sentence to:

The following operations manipulate the DWARF stack, and may operate on both values and locations.... [remainder of paragraph unchanged]

Change the second paragraph to:

Each entry on the stack is either a value (with an associated type) or a location.

Include the descriptions for the following operations:

For the above operations, remove all occurrences of "including its type identifier".

For DW_OP_dup, change the description to:

The DW_OP_dup operation duplicates the entry at the top of the stack.

For DW_OP_drop, change the description to:

The DW_OP_drop operation pops the entry at the top of the stack.

The following operations that were in section 2.5.2.3 are moved to other sections:

Section 3.3 Literal and Constant Operations [was 2.5.2.1 Literal Encodings]

Rename and place the contents of old section 2.5.2.1 here.

Include the descriptions of the following operations:

For DW_OP_constx, change "size of a machine address" to "size of the generic type."

For DW_OP_const_type, insert "a" after "The second operand is".

The following operations that were in 2.5.2.1 are moved to Section 3.7 Memory Locations:

Section 3.4 Register Value Operations [was 2.5.2.2]

Place the contents of old section 2.5.2.2 here.

Replace the first paragraph with the following:

The following operation pushes the contents of a register onto the stack.

Include the descriptions of the following operations:

Remove the DW_OP_regval_bits operator, which was added in DWARF 6 by Issue 201007.1. With the ability to dereference register locations, this operator is not needed.

The following operations that were in 2.5.2.2 are moved to other sections:

Section 3.5 Arithmetic and Logical Operations [was 2.5.2.4]

Place the contents of old section 2.5.2.4 here.

Remove the second paragraph:

If the type of the operands is the generic type, except as otherwise specified, the arithmetic operations perform addressing arithmetic, that is, unsigned arithmetic that is performed modulo one plus the largest representable address.

Include the descriptions of the following operations:

Section 3.6 Context Query Operations [NEW]

Insert the following into this new section:

The following operations can be used to push a value or location obtained from the expression evaluation context (see Section 3.1) onto the stack:

  1. DW_OP_push_object_location [moved from section 2.5.2.3 and renamed]
    The DW_OP_push_object_location operation pushes the location of the current object (see section 3.1) onto the stack, as part of evaluation of a user-presented expression.

This object may correspond to an independent variable described by its own debugging information entry; or it may be a component of an array, structure, or class whose address has been dynamically determined by an earlier step during user expression evaluation.

This operator provides explicit functionality (especially for arrays involving descriptors) that is analogous to the implicit push of the base address of a structure prior to evaluation of a DW_AT_data_member_location to access a data member of a structure. For an example, see Appendix D.2 on page 304.

In previous versions of DWARF, this operator was named DW_OP_push_object_address. The old name is still supported in DWARF 6 for compatibility.

  1. DW_OP_form_tls_location [moved from section 2.5.2.3] The DW_OP_form_tls_location operation pops a value from the stack, which must have an integral type, translates this value into a location in the thread-local storage for the current thread (see Section 3.1), and pushes the location onto the stack. The meaning of the value.... [remainder of paragraph unchanged]

Some implementations of C, C++, Fortran, and other languages, support a thread-local storage class. Variables with this storage class have distinct values and addresses in distinct threads, much as automatic variables have distinct values and addresses in each function invocation. Typically, there is a single block of storage containing all thread-local variables declared in the main executable, and a separate block for the variables declared in each shared library. Each thread-local variable can then be accessed in its block using an identifier. This identifier is typically an offset into the block and pushed onto the DWARF stack by one of the DW_OP_const<n><x> operations prior to the DW_OP_form_tls_location operation. Computing the address of the appropriate block can be complex (in some cases, the compiler emits a function call to do it), and difficult to describe using ordinary DWARF location expressions. Instead of forcing complex thread-local storage calculations into the DWARF expressions, the DW_OP_form_tls_location operation allows the consumer to perform the computation based on the run-time environment.

In previous versions of DWARF, this operator was named DW_OP_form_tls_address. The old name is still supported in DWARF 6 for compatibility.

  1. DW_OP_call_frame_cfa... [moved unchanged from section 2.5.2.3]

  2. DW_OP_push_lane... [moved unchanged from section 2.5.2.3]

Elsewhere in the document, change DW_OP_push_object_address to DW_OP_push_object_location, and DW_OP_form_tls_address to DW_OP_form_tls_location

Section 3.7 Memory Locations [adapted from 2.6.1.1.2]

Insert the following:

A memory location represents the location of a piece or all of an object or other entity in memory. On architectures that support multiple address spaces, a memory location contains a component that identifies the address space.

In contexts that expect a location, a value of the generic type will be implicitly converted to a memory location in the default address space.

The following operations push memory locations onto the stack:

  1. DW_OP_addr [moved from section 2.5.2.1]
    The DW_OP_addr operation has a single operand that encodes a machine address and whose size is the size of an address on the target machine. The value of this operand is treated as an address in the default address space and the corresponding memory location is pushed onto the stack.

  2. DW_OP_addrx [moved from section 2.5.2.1]
    The DW_OP_addrx operation has a single operand that encodes an unsigned LEB128 value, which is a zero-based index into the .debug_addr section, where a machine address is stored. This index is relative to the value of the DW_AT_addr_base attribute of the associated compilation unit. The address obtained is treated as an address in the default address space and the corresponding memory location is pushed onto the stack.

  3. DW_OP_fbreg... [moved from section 2.5.2.2]
    The DW_OP_fbreg operation provides a signed LEB128 byte offset B from the location specified by the location expression in the DW_AT_frame_base attribute of the current function (see Section 3.1). The frame base location, offset by B bytes, is pushed onto the stack.

This is typically a stack pointer register plus or minus some offset.

  1. DW_OP_breg0, ..., DW_OP_breg31 [moved from section 2.5.2.2]
    The single operand of the DW_OP_breg<n> operations provides a signed LEB128 byte offset. The contents of the specified register (0–31) are treated as a memory address in the default address space. The offset is added to the address obtained from the register and the resulting memory location is pushed onto the stack.

  2. DW_OP_bregx [moved from section 2.5.2.2]
    The DW_OP_bregx operation has two operands. The first operand is a register number which is specified by an unsigned LEB128 number. The second operand is a signed LEB128 byte offset. It is the same as DW_OP_breg<n> except it uses the register and offset provided by the operands.

Section 3.8 Register Locations [adapted from 2.6.1.1.3]

Place the contents of old section 2.6.1.1.3 here.

Remove the first paragraph:

A register location consists of a register name operation, which represents a piece or all of an object located in a given register.

Remove the last sentence of non-normative text that follows:

A register location description must stand alone as the entire description of an object or a piece of an object.

Include the descriptions of the following operations:

For DW_OP_reg<n>, replace

The object addressed is in register n.

with:

A location descriptor that names the designated register, with an offset of 0, is formed and pushed on the stack.

For DW_OP_regx, add the sentence:

A location descriptor that names the designated register, with an offset of 0, is formed and pushed on the stack.

Replace the non-normative paragraph at the end with the following:

These operations name a register, not the contents of the register. To fetch the contents of a register, it is necessary to use one of the register based addressing operations, such as DW_OP_bregx (Section 3.7), a register value operation, such as DW_OP_regval_type (Section 3.4), or a DW_OP_reg operation followed by a derefencing operation (Section 3.12).

Section 3.9 Undefined Locations [adapted from 2.6.1.1.1]

Insert the following (adapted from Section 2.6.1.1.1):

An undefined location represents a piece or all of an object that is present in the source but not in the object code (perhaps due to optimization). An undefined location cannot be read from or written to.

  1. DW_OP_undefined
    The DW_OP_undefined operation pushes an undefined location with an offset of 0 onto the stack.

  2. A DWARF expression containing no operations or that leaves no elements on the stack also produces an undefined location. This is for compatibility with earlier versions of DWARF.

Section 3.10 Implicit Locations [adapted from 2.6.1.1.4]

Move the contents of Section 2.6.1.1.4 here.

Move the last non-normative paragraph to the start of the section and replace the term "location descriptions" with "location expressions":

DWARF location expressions are intended ...

In the remainder of the section, replace the term "location description" with "location" throughout.

In the first (normative) paragraph, change "but whose contents are nonetheless either known or known to be undefined" to "but whose contents are nonetheless known".

Include the descriptions of the following operations:

For DW_OP_implicit_value, add:

A location descriptor L is formed for a block of implicit storage which contains the given byte sequence. The offset of L is set to 0, and L is pushed onto the stack.

For DW_OP_stack_value, replace:

In this form of location description, the DWARF expression represents the actual value of the object, rather than its location. The DW_OP_stack_value operation terminates the expression.

with:

The value V on top of the stack is popped, and a location descriptor for L is formed for a block of implicit storage containing the value V, represented using the encoding and byte order of the value's type. The offset of L is set to 0, and L is pushed onto the stack.

For DW_OP_implicit_pointer, in the first (non-normative) paragraph, replace "this value" with "the latter value".

For DW_OP_implicit_pointer, replace the 3rd paragraph:

The DW_OP_implicit_pointer operation has two operands: a reference to a debugging information entry that describes the dereferenced object's value, and a signed number that is treated as a byte offset from the start of that value. The first operand is a 4-byte unsigned value in the 32-bit DWARF format, or an 8-byte unsigned value in the 64-bit DWARF format (see Section {32bitand64bitdwarfformats}) that is used as the offset of a debugging information entry in the .debug_info section of the current executable or shared object file. The second operand is a signed LEB128 number.

with:

The DW_OP_implicit_pointer operation has two operands: a reference to a debugging information entry D that describes the dereferenced object's value, and a signed number that is treated as a byte offset B from the start of that value. The first operand is a 4-byte unsigned value in the 32-bit DWARF format, or an 8-byte unsigned value in the 64-bit DWARF format (see Section {32bitand64bitdwarfformats}) that is used as the offset of the debugging information entry D in the .debug_info section of the current executable or shared object file. The second operand, B, is a signed LEB128 number.

And add:

An implicit storage location LP is created for the location of the pointer object P, and the location LV of the object V is obtained from D. LP describes the location of P, with an offset of 0, and the value in that implicit storage represents (implicitly) the value of the pointer, which is meant to point to V. LV describes the location of the object V, with a byte offset of B.

The location LP is pushed onto the stack.

In the non-normative paragraph that follows, change "The debugging information entry referenced by..." to "The debugging information entry D referenced by...."

Section 3.11 Composite Locations [adapted from 2.6.1.2]

Insert the following (adapted from Section 2.6.1.2, and with the new DW_OP_composite operator):

A composite location descriptor represents the location of an object or value that does not exist in a single block of contiguous storage (e.g., as the result of compiler optimization where part of the object is promoted to a register). It consists of a set of tuples (as described in Section 2.5), where each tuple describes a range of bits from the object and holds a location descriptor for that range of bits.

The piece operations (DW_OP_piece and DW_OP_bit_piece) in this section are used to build a composite location descriptor in storage order, one piece at a time. Each piece operation expects a location descriptor LP at the top of the stack, and a composite location descriptor LC in the preceding element on the stack. It pops those two location descriptors and extends LC by adding a new mapping tuple (SP, EP, LP), where SP is equal to the largest value of E from among the existing mapping tuples of LC, and EP is determined by the size operand of the piece operation. The extended location descriptor LC is pushed onto the stack. When adding a piece to an empty composite, SP is 0.

  1. DW_OP_composite

    The DW_OP_composite operator has no operands. It pushes a new, empty, composite location descriptor onto the stack, with an offset of 0.

    This operator is provided so that a new series of piece operations can be started to form a composite location when the state of the stack is unknown (e.g., following a DW_OP_call operation), or when a new composite is to be started (e.g., rather than add to a previous composite location on the stack).

  2. DW_OP_piece

    The DW_OP_piece operation takes a single operand, which is an unsigned LEB128 number. The number describes the size N, in bytes, of the piece of the object to be appended to the composite LC. The EP value of the new tuple is set to SP + N × 8.

    If LP is a register location, but the piece does not occupy the entire register, the placement of the piece within that register is defined by the ABI.

    Many compilers store a single variable in sets of registers, or store a variable partially in memory and partially in registers. DW_OP_piece provides a way of describing how large a part of a variable a particular location refers to.

  3. DW_OP_bit_piece

    The DW_OP_bit_piece operation takes two operands. The first is an unsigned LEB128 number that gives the size N, in bits, of the piece to be appended. The second is an unsigned LEB128 number that gives the offset in bits to be applied to the location LP. The EP value of the new tuple is set to SP + L.

    Interpretation of the offset depends on the type of location. If the location is an undefined location (see Section 3.10), the DW_OP_bit_piece operation describes a piece consisting of the given number of bits whose values are undefined, and the offset is ignored. If the location is a memory location (see Section 3.7), the DW_OP_bit_piece operation describes a sequence of bits relative to the location whose address is on the top of the DWARF stack using the bit numbering and direction conventions that are appropriate to the current language on the target system. In all other cases, the source of the piece is given by either a register location (see Section 3.8) or an implicit value location (see Section 3.9); the offset is from the least significant bit of the source value.

    The DW_OP_bit_piece operator is used instead of DW_OP_piece when the piece to be assembled into a value or assigned to is not byte-sized or is not at the start of a register or addressable unit of memory.

    Whether or not a DW_OP_piece operation is equivalent to any DW_OP_bit_piece operation with an offset of 0 is ABI dependent.

For compatibility with DWARF Version 5 and earlier, the following additional rules apply to piece operations:

  • If LP is a non-composite location (or convertible to one) and is the only element on the stack, the result is a new composite location with the single piece LP (as if DW_OP_composite DW_OP_swap had been processed immediately prior to the piece operation). This rule supports the first piece operation in a DWARF 5 expression.

  • If a piece operation is processed while the stack is empty, a new empty composite and an undefined location are pushed implicitly (as if DW_OP_composite DW_OP_undefined had been processed immediately prior to the piece operation). The result is a composite with a single undefined piece. This rule supports the empty piece operation in DWARF 5 when it is the first piece of a composite.

  • If the top of the stack is a composite location, and is the only element on the stack, an undefined location is pushed implicitly (as if DW_OP_undefined had been processed immediately prior to the piece operation), whereupon the composite on top of the stack is taken as LC and the undefined location is LP. The result is the addition of an undefined piece to the existing composite location. This rule supports the empty piece operation in DWARF 5 when it is not the first piece of a composite.

  • Otherwise, the DWARF expression is not valid.

Section 3.12 Dereferencing Operations [NEW]

Add the following introductory paragraphs:

The following operations are used to dereference a location on the stack; i.e., to load a value (or another location) stored at that location. Each operation defines the number of bytes or bits to be loaded.

Dereferencing a memory location simply loads a value from memory. Dereferencing a register location extracts all or some of the bits from the register, depending on the offset of the location and the size of value to be loaded. Dereferencing an implicit location loads a value (or, in the case of an implicit pointer, another location) from the implicit storage. Attempting to dereference an undefined location is an error.

When dereferencing from a composite location, the value to be loaded may span more than one of the pieces of the composite, and the consumer must reassemble the value from the component pieces.

Include the descriptions for the following operations (moved from 2.5.2.3):

For DW_OP_deref, change the description to:

The DW_OP_deref operation pops a location L from the top of the stack. The first S bits, where S is the size (in bits) of an address on the target machine, are retrieved from the location L and pushed onto the stack as a value of the generic type.

If the location on the stack is an implicit pointer (see Section 3.10), the location LP is dereferenced to obtain the location LV. In this case, the offset of the location LP must be 0.

For DW_OP_deref_size, change the description to:

The DW_OP_deref_size takes a single 1-byte unsigned integral operand that specifies the size S, in bytes, of the value to be retrieved. The size S must be no larger than the size of the generic type. The operation behaves like the DW_OP_deref operation: it pops a location L from the stack. The first S bytes are retrieved from the location L, zero extended to the size of the generic type, and pushed onto the stack as a value of the generic type.

For DW_OP_deref_type, change the description to:

The DW_OP_deref_type operation takes two operands. The first operand is a 1-byte unsigned integer that specifies the byte size S of the type given by the second operand. The second operand is an unsigned LEB128 integer that represents the offset of a debugging information entry in the current compilation unit, which must be a DW_TAG_base_type entry that provides the type T of the value to be retrieved. The size S must be the same as the byte size of the base type represented by the type T. This operation pops a location L from the stack. The first S bytes are retrieved from the location L and pushed onto the stack as a value of type T.

And delete the non-normative paragraph following (this rationale is appropriate for DW_OP_const_type, but not for DW_OP_deref_type):

While the size of the pushed value could be inferred from the base type definition, it is encoded explicitly into the operation so that the operation can be parsed easily without reference to the .debug_info section.

For DW_OP_xderef and DW_OP_xderef_size, change "integral type identifiers" to "integral types," and "generic type identifier" to "generic type."

For DW_OP_xderef_type, change "whose value value which is" to "whose value is". [This was a typo in the DWARF 5 spec.]

Section 3.13 Offset Operations [NEW]

Add:

In addition to the composite operations, locations may be modified by the following operations:

  1. DW_OP_offset

    DW_OP_offset pops two stack entries. The first (top of stack) must be an integral type value, which represents a signed byte displacement. The second must be a location. It forms an updated location by adding the given byte displacement to the offset component of the original location and pushes the updated location onto the stack.

  2. DW_OP_bit_offset

    DW_OP_bit_offset pops two stack entries. The first (top of stack) must be an integral type value, which represents a signed bit displacement. The second must be a location. It forms an updated location by adding the given bit displacement to the offset component of the original location and pushes the updated location onto the stack.

    A bit offset of n*8 is equivalent to a byte offset of n.

The resulting offset must be within range of the location's storage bank.

Section 3.14 Control Flow Operations [was 2.5.2.5]

Move the contents of old section 2.5.2.5 here.

Include the descriptions of the following operations:

For DW_OP_bra, change:

If the value popped is not the constant 0, the 2-byte constant operand is ...

to

If the value popped is not 0, the constant operand is ...

Under DW_OP_call2, etc., change:

Execution of the DWARF expression of a DW_AT_location attribute may add to and/or remove from values on the stack. Execution returns to the point following the call when the end of the attribute is reached. Values on the stack at the time of the call may be used as parameters by the called expression and values left on the stack by the called expression may be used as return values by prior agreement between the calling and called expressions.

to:

Execution of the DWARF expression of a DW_AT_location attribute may pop elements from the stack and/or push values or locations onto the stack. Execution returns to the point following the call when the end of the attribute is reached. Values and locations on the stack at the time of the call may be used as parameters by the called expression, and elements (values or locations) left on the stack by the called expression may be used as return values by prior agreement between the calling and called expressions.

Section 3.15 Type Conversions [was 2.5.2.6]

Move and renumber Section 2.5.2.6 to here.

Include the descriptions of the following operations:

Under DW_OP_reinterpret, in the next-to-last sentence, change "converted" to "reinterpreted".

Section 3.16 Special Operations [was 2.5.2.7]

Move and renumber Section 2.5.2.7 to here.

Include the descriptions of the following operations:

For DW_OP_nop, change "location stack" to "stack" and "values" to "elements".

For DW_OP_user_extended, in the last (non-normative) paragraph, change "the standard" to "this specification".

Section 3.17 Value Lists [was 2.5.2]

Place the contents of old section 2.5.2 Value Lists here.

Remove the fifth (non-normative) paragraph:

The DWARF expressions in value list entries, being expressions and not location descriptions, may not contain any of the DWARF operations described in Section {locationdescriptions}.

Section 3.18 Location Lists [was 2.6.2]

Place the contents of old Section 2.6.2 Location Lists here.

Change the first sentence to:

Location lists are used in place of location expressions whenever the object whose location is being described can change location during its lifetime.

In the second paragraph, change "a location or other attribute" to "an attribute."

Remove the third (non-normative) paragraph.

Section 5.7.6 Data Member Entries

In the description for DW_AT_data_member_location, change the first paragraph of the second bullet to:

2. Otherwise, the value must be a location expression. The location of the containing entity is pushed on the DWARF stack before the location expression is evaluated; the result of the evaluation is the location of the member entry (see Section 3.1).

Replace the following non-normative paragraph with:

The push on the DWARF expression stack of the location of the containing construct is equivalent to execution of the DW_OP_push_object_location operation (see Section 3.6); DW_OP_push_object_location therefore is not needed at the beginning of a location expression for a data member. The result of the evaluation is a location, not an offset to the member.

Remove the second non-normative paragraph:

A DW_AT_data_member_location attribute that has the form of a location description is not valid for a data member contained in an entity that is not byte aligned because DWARF operations do not allow for manipulating or computing bit offsets.

Section 5.14 Pointer to Member Type Entries

Replace the paragraph beginning "The DW_AT_use_location description..." with:

The DW_AT_use_location expression is used in conjunction with the location expressions for a particular object of the given pointer to member type and for a particular structure or class instance. The DW_AT_use_location attribute expects two values to be pushed onto the DWARF expression stack before the DW_AT_use_location expression is evaluated. The first value pushed is the value of the pointer to member object itself. The second value pushed is the location of the entire structure or union instance containing the member whose location is being calculated.

Section 6.4.1 Structure of Call Frame Information

For the "expression(E)" register rule, change the description to:

The previous value of this register is located at the location produced by executing the DWARF expression E (see Chapter 3).

Section 7.5.5 Classes and Forms

Replace the locdesc bullet with the following:

  • locexpr
    A DWARF location expression (see Section 2.5). This is represented as an unsigned LEB128 length, followed by a byte sequence of the specified length (DW_FORM_locexpr) containing the location expression.

Rename DW_FORM_exprloc to DW_FORM_expression.

Section 7.7 DWARF Expressions and Location Descriptions

Rename Section 7.7 to "DWARF Expressions."

Section 7.7.1 DWARF Expressions

Add to Table 7.9:

                            No. of
Operation             Code  Operands  Notes
--------------------  ----  --------  -----
DW_OP_offset          TBA      0
DW_OP_bit_offset      TBA      0
DW_OP_composite       TBA      0
DW_OP_undefined       TBA      0

Section 7.7.2 Location Descriptions

Remove Section 7.7.2.

Section D.1.3 DWARF Location Expression Examples [was DWARF Location Description Examples]

Change the name of the section. Also change "location descriptions" to "location expressions" in the first paragraph.

Replace the seventh example (DW_OP_plus_uconst) with:

DW_OP_lit4
DW_OP_offset

A structure member is four bytes from the start of the structure instance. The base location is assumed to be already on the stack.

In the four composite examples that follow, add a DW_OP_composite operator at the beginning of each sequence, and change DW_OP_plus to DW_OP_offset.

In the sixth DW_OP_entry_value example, replace DW_OP_plus_uconst (16) with DW_OP_lit16 and DW_OP_offset.

Section D.2.1 Fortran Simple Array Example

In Figure D.4, change DW_OP_plus to DW_OP_offset in the following places:

Section D.2.3 Fortran 2008 Assumed-rank Array Example

[Page 302] In Figure D.13, change DW_OP_plus to DW_OP_offset in the following places:

2023-05-24: Original proposal.

2025-02-28: Rewritten.

2025-03-28: Revised. Incorporated review comments.

2025-04-11: Revised. Rewrote parts of "Values and Locations," cleaned up leftover uses of "location description," changed some more examples in Appendix D.

dwarfstd.org is supported by Sourceware. Contributions are welcome.

Copyright © 2023–2025 DWARF Debugging Information Format Committee

This work is licensed under CC BY 4.0