Debian Codegraph @

llvm/docs/AMDGPUDwarfExtensionAllowLocationDescriptionOnTheDwarfExpressionStack/AMDGPUDwarfExtensionAllowLocationDescriptionOnTheDwarfExpressionStack.md

raw
1 # Allow Location Descriptions on the DWARF Expression Stack <!-- omit in toc -->
2
3 ```{contents}
4 ---
5 local:
6 ---
7 ```
8
9 # 1. Extension
10
11 In DWARF 5, expressions are evaluated using a typed value stack, a separate
12 location area, and an independent loclist mechanism. This extension unifies all
13 three mechanisms into a single generalized DWARF expression evaluation model
14 that allows both typed values and location descriptions to be manipulated on the
15 evaluation stack. Both single and multiple location descriptions are supported
16 on the stack. In addition, the call frame information (CFI) is extended to
17 support the full generality of location descriptions. This is done in a manner
18 that is backwards compatible with DWARF 5. The extension involves changes to the
19 DWARF 5 sections 2.5 (pp 26-38), 2.6 (pp 38-45), and 6.4 (pp 171-182).
20
21 The extension permits operations to act on location descriptions in an
22 incremental, consistent, and composable manner. It allows a small number of
23 operations to be defined to address the requirements of heterogeneous devices as
24 well as providing benefits to non-heterogeneous devices. It acts as a foundation
25 to provide support for other issues that have been raised that would benefit all
26 devices.
27
28 Other approaches were explored that involved adding specialized operations and
29 rules. However, these resulted in the need for more operations that did not
30 compose. It also resulted in operations with context sensitive semantics and
31 corner cases that had to be defined. The observation was that numerous
32 specialized context sensitive operations are harder for both producers and
33 consumers than a smaller number of general composable operations that have
34 consistent semantics regardless of context.
35
36 First, section [2. Heterogeneous Computing Devices](#heterogeneous-computing-devices)
37 describes heterogeneous devices and the features they have that are not addressed by DWARF 5.
38 Then section [3. DWARF5](#dwarf-5) presents a brief simplified overview of the DWARF 5 expression
39 evaluation model that highlights the difficulties for supporting the
40 heterogeneous features. Next, section [4. Extension
41 Solution](#extension-solution) provides an overview of the proposal, using
42 simplified examples to illustrate how it can address the issues of heterogeneous
43 devices and also benefit non-heterogeneous devices. Then overall conclusions are
44 covered in section [5. Conclusion](#conclusion).
45 Appendix [A. Changes to DWARF Debugging Information Format Version 5](#changes-to-dwarf-debugging-information-format-version-5) gives changes
46 relative to the DWARF Version 5 standard. Finally, appendix
47 [B. Further Information](#further-information) has references to further information.
48
49 # 2. Heterogeneous Computing Devices
50
51 GPUs and other heterogeneous computing devices have features not common to CPU
52 computing devices.
53
54 These devices often have many more registers than a CPU. This helps reduce
55 memory accesses which tend to be more expensive than on a CPU due to the much
56 larger number of threads concurrently executing. In addition to traditional
57 scalar registers of a CPU, these devices often have many wide vector registers.
58
59 ![Example GPU Hardware](images/example-gpu-hardware.png)
60
61 They may support masked vector instructions that are used by the compiler to map
62 high level language threads onto the lanes of the vector registers. As a
63 consequence, multiple language threads execute in lockstep as the vector
64 instructions are executed. This is termed single instruction multiple thread
65 (SIMT) execution.
66
67 ![SIMT/SIMD Execution Model](images/simt-execution-model.png)
68
69 GPUs can have multiple memory address spaces in addition to the single global
70 memory address space of a CPU. These additional address spaces are accessed
71 using distinct instructions and are often local to a particular thread or group
72 of threads.
73
74 For example, a GPU may have a per thread block address space that is implemented
75 as scratch pad memory with explicit hardware support to isolate portions to
76 specific groups of threads created as a single thread block.
77
78 A GPU may also use global memory in a non linear manner. For example, to support
79 providing a SIMT per lane address space efficiently, there may be instructions
80 that support interleaved access.
81
82 Through optimization, the source variables may be located across these different
83 storage kinds. SIMT execution requires locations to be able to express selection
84 of runtime defined pieces of vector registers. With the more complex locations,
85 there is a benefit to be able to factorize their calculation which requires all
86 location kinds to be supported uniformly, otherwise duplication is necessary.
87
88 # 3. DWARF 5
89
90 Before presenting the proposed solution to supporting heterogeneous devices, a
91 brief overview of the DWARF 5 expression evaluation model will be given to
92 highlight the aspects being addressed by the extension.
93
94 ## 3.1 How DWARF Maps Source Language To Hardware
95
96 DWARF is a standardized way to specify debug information. It describes source
97 language entities such as compilation units, functions, types, variables, etc.
98 It is either embedded directly in sections of the code object executables, or
99 split into separate files that they reference.
100
101 DWARF maps between source program language entities and their hardware
102 representations. For example:
103
104 - It maps a hardware instruction program counter to a source language program
105 line, and vice versa.
106 - It maps a source language function to the hardware instruction program counter
107 for its entry point.
108 - It maps a source language variable to its hardware location when at a
109 particular program counter.
110 - It provides information to allow virtual unwinding of hardware registers for a
111 source language function call stack.
112 - In addition, it provides numerous other information about the source language
113 program.
114
115 In particular, there is great diversity in the way a source language entity
116 could be mapped to a hardware location. The location may involve runtime values.
117 For example, a source language variable location could be:
118
119 - In register.
120 - At a memory address.
121 - At an offset from the current stack pointer.
122 - Optimized away, but with a known compiler time value.
123 - Optimized away, but with an unknown value, such as happens for unused
124 variables.
125 - Spread across combination of the above kinds of locations.
126 - At a memory address, but also transiently loaded into registers.
127
128 To support this DWARF 5 defines a rich expression language comprised of loclist
129 expressions and operation expressions. Loclist expressions allow the result to
130 vary depending on the PC. Operation expressions are made up of a list of
131 operations that are evaluated on a simple stack machine.
132
133 A DWARF expression can be used as the value of different attributes of different
134 debug information entries (DIE). A DWARF expression can also be used as an
135 argument to call frame information information (CFI) entry operations. An
136 expression is evaluated in a context dictated by where it is used. The context
137 may include:
138
139 - Whether the expression needs to produce a value or the location of an entity.
140 - The current execution point including process, thread, PC, and stack frame.
141 - Some expressions are evaluated with the stack initialized with a specific
142 value or with the location of a base object that is available using the
143 DW_OP_push_object_address operation.
144
145 ## 3.2 Examples
146
147 The following examples illustrate how DWARF expressions involving operations are
148 evaluated in DWARF 5. DWARF also has expressions involving location lists that
149 are not covered in these examples.
150
151 ### 3.2.1 Dynamic Array Size
152
153 The first example is for an operation expression associated with a DIE attribute
154 that provides the number of elements in a dynamic array type. Such an attribute
155 dictates that the expression must be evaluated in the context of providing a
156 value result kind.
157
158 ![Dynamic Array Size Example](images/01-value.example.png)
159
160 In this hypothetical example, the compiler has allocated an array descriptor in
161 memory and placed the descriptor's address in architecture register SGPR0. The
162 first location of the array descriptor is the runtime size of the array.
163
164 A possible expression to retrieve the dynamic size of the array is:
165
166 DW_OP_regval_type SGPR0 Generic
167 DW_OP_deref
168
169 The expression is evaluated one operation at a time. Operations have operands
170 and can pop and push entries on a stack.
171
172 ![Dynamic Array Size Example: Step 1](images/01-value.example.frame.1.png)
173
174 The expression evaluation starts with the first DW_OP_regval_type operation.
175 This operation reads the current value of an architecture register specified by
176 its first operand: SGPR0. The second operand specifies the size of the data to
177 read. The read value is pushed on the stack. Each stack element is a value and
178 its associated type.
179
180 ![Dynamic Array Size Example: Step 2](images/01-value.example.frame.2.png)
181
182 The type must be a DWARF base type. It specifies the encoding, byte ordering,
183 and size of values of the type. DWARF defines that each architecture has a
184 default generic type: it is an architecture specific integral encoding and byte
185 ordering, that is the size of the architecture's global memory address.
186
187 The DW_OP_deref operation pops a value off the stack, treats it as a global
188 memory address, and reads the contents of that location using the generic type.
189 It pushes the read value on the stack as the value and its associated generic
190 type.
191
192 ![Dynamic Array Size Example: Step 3](images/01-value.example.frame.3.png)
193
194 The evaluation stops when it reaches the end of the expression. The result of an
195 expression that is evaluated with a value result kind context is the top element
196 of the stack, which provides the value and its type.
197
198 ### 3.2.2 Variable Location in Register
199
200 This example is for an operation expression associated with a DIE attribute that
201 provides the location of a source language variable. Such an attribute dictates
202 that the expression must be evaluated in the context of providing a location
203 result kind.
204
205 DWARF defines the locations of objects in terms of location descriptions.
206
207 In this example, the compiler has allocated a source language variable in
208 architecture register SGPR0.
209
210 ![Variable Location in Register Example](images/02-reg.example.png)
211
212 A possible expression to specify the location of the variable is:
213
214 DW_OP_regx SGPR0
215
216 ![Variable Location in Register Example: Step 1](images/02-reg.example.frame.1.png)
217
218 The DW_OP_regx operation creates a location description that specifies the
219 location of the architecture register specified by the operand: SGPR0. Unlike
220 values, location descriptions are not pushed on the stack. Instead they are
221 conceptually placed in a location area. Unlike values, location descriptions do
222 not have an associated type, they only denote the location of the base of the
223 object.
224
225 ![Variable Location in Register Example: Step 2](images/02-reg.example.frame.2.png)
226
227 Again, evaluation stops when it reaches the end of the expression. The result of
228 an expression that is evaluated with a location result kind context is the
229 location description in the location area.
230
231 ### 3.2.3 Variable Location in Memory
232
233 The next example is for an operation expression associated with a DIE attribute
234 that provides the location of a source language variable that is allocated in a
235 stack frame. The compiler has placed the stack frame pointer in architecture
236 register SGPR0, and allocated the variable at offset 0x10 from the stack frame
237 base. The stack frames are allocated in global memory, so SGPR0 contains a
238 global memory address.
239
240 ![Variable Location in Memory Example](images/03-memory.example.png)
241
242 A possible expression to specify the location of the variable is:
243
244 DW_OP_regval_type SGPR0 Generic
245 DW_OP_plus_uconst 0x10
246
247 ![Variable Location in Memory Example: Step 1](images/03-memory.example.frame.1.png)
248
249 As in the previous example, the DW_OP_regval_type operation pushes the stack
250 frame pointer global memory address onto the stack. The generic type is the size
251 of a global memory address.
252
253 ![Variable Location in Memory Example: Step 2](images/03-memory.example.frame.2.png)
254
255 The DW_OP_plus_uconst operation pops a value from the stack, which must have a
256 type with an integral encoding, adds the value of its operand, and pushes the
257 result back on the stack with the same associated type. In this example, that
258 computes the global memory address of the source language variable.
259
260 ![Variable Location in Memory Example: Step 3](images/03-memory.example.frame.3.png)
261
262 Evaluation stops when it reaches the end of the expression. If the expression
263 that is evaluated has a location result kind context, and the location area is
264 empty, then the top stack element must be a value with the generic type. The
265 value is implicitly popped from the stack, and treated as a global memory
266 address to create a global memory location description, which is placed in the
267 location area. The result of the expression is the location description in the
268 location area.
269
270 ![Variable Location in Memory Example: Step 4](images/03-memory.example.frame.4.png)
271
272 ### 3.2.4 Variable Spread Across Different Locations
273
274 This example is for a source variable that is partly in a register, partly undefined, and partly in memory.
275
276 ![Variable Spread Across Different Locations Example](images/04-composite.example.png)
277
278 DWARF defines composite location descriptions that can have one or more parts.
279 Each part specifies a location description and the number of bytes used from it.
280 The following operation expression creates a composite location description.
281
282 DW_OP_regx SGPR3
283 DW_OP_piece 4
284 DW_OP_piece 2
285 DW_OP_bregx SGPR0 0x10
286 DW_OP_piece 2
287
288 ![Variable Spread Across Different Locations Example: Step 1](images/04-composite.example.frame.1.png)
289
290 The DW_OP_regx operation creates a register location description in the location
291 area.
292
293 ![Variable Spread Across Different Locations Example: Step 2](images/04-composite.example.frame.2.png)
294
295 The first DW_OP_piece operation creates an incomplete composite location
296 description in the location area with a single part. The location description in
297 the location area is used to define the beginning of the part for the size
298 specified by the operand, namely 4 bytes.
299
300 ![Variable Spread Across Different Locations Example: Step 3](images/04-composite.example.frame.3.png)
301
302 A subsequent DW_OP_piece adds a new part to an incomplete composite location
303 description already in the location area. The parts form a contiguous set of
304 bytes. If there are no other location descriptions in the location area, and no
305 value on the stack, then the part implicitly uses the undefined location
306 description. Again, the operand specifies the size of the part in bytes. The
307 undefined location description can be used to indicate a part that has been
308 optimized away. In this case, 2 bytes of undefined value.
309
310 ![Variable Spread Across Different Locations Example: Step 4](images/04-composite.example.frame.4.png)
311
312 The DW_OP_bregx operation reads the architecture register specified by the first
313 operand (SGPR0) as the generic type, adds the value of the second operand
314 (0x10), and pushes the value on the stack.
315
316 ![Variable Spread Across Different Locations Example: Step 5](images/04-composite.example.frame.5.png)
317
318 The next DW_OP_piece operation adds another part to the already created
319 incomplete composite location.
320
321 If there is no other location in the location area, but there is a value on
322 stack, the new part is a memory location description. The memory address used is
323 popped from the stack. In this case, the operand of 2 indicates there are 2
324 bytes from memory.
325
326 ![Variable Spread Across Different Locations Example: Step 6](images/04-composite.example.frame.6.png)
327
328 Evaluation stops when it reaches the end of the expression. If the expression
329 that is evaluated has a location result kind context, and the location area has
330 an incomplete composite location description, the incomplete composite location
331 is implicitly converted to a complete composite location description. The result
332 of the expression is the location description in the location area.
333
334 ![Variable Spread Across Different Locations Example: Step 7](images/04-composite.example.frame.7.png)
335
336 ### 3.2.5 Offsetting a Composite Location
337
338 This example attempts to extend the previous example to offset the composite
339 location description it created. The [3.2.3 Variable Location in
340 Memory](#variable-location-in-memory) example conveniently used the DW_OP_plus
341 operation to offset a memory address.
342
343 DW_OP_regx SGPR3
344 DW_OP_piece 4
345 DW_OP_piece 2
346 DW_OP_bregx SGPR0 0x10
347 DW_OP_piece 2
348 DW_OP_plus_uconst 5
349
350 ![Offsetting a Composite Location Example: Step 6](images/05-composite-plus.example.frame.1.png)
351
352 However, DW_OP_plus cannot be used to offset a composite location. It only
353 operates on the stack.
354
355 ![Offsetting a Composite Location Example: Step 7](images/05-composite-plus.example.frame.2.png)
356
357 To offset a composite location description, the compiler would need to make a
358 different composite location description, starting at the part corresponding to
359 the offset. For example:
360
361 DW_OP_piece 1
362 DW_OP_bregx SGPR0 0x10
363 DW_OP_piece 2
364
365 This illustrates that operations on stack values are not composable with
366 operations on location descriptions.
367
368 ### 3.2.6 Pointer to Member
369
370 > NOTE: Without loss of generality, DWARF 4 is used in this example as full
371 > support for DWARF 5 is not present in the versions of the tools used. No
372 > feature of DWARF 5 provides a remedy for this issue.
373
374 This example highlights the inability of DWARF 5 to describe C++
375 pointer-to-member use semantics.
376
377 The mechanism DWARF 5 provides for describing pointer-to-member use is
378 `DW_AT_use_location`, which is defined as encoding a location description which
379 computes the address of the member pointed to by a pointer-to-member, given the
380 pointer-to-member object and the address of the containing object.
381
382 That is, when a debug agent wishes to evaluate a pointer-to-member access
383 operation, it first pushes two values onto the DWARF expression stack:
384
385 * The pointer-to-member object
386 * The address of the containing object
387
388 It then evaluates the location description associated with the
389 `DW_AT_use_location` of the pointer-to-member type, and interprets the result
390 as the address of the member pointed to by the pointer-to-member.
391
392 Consider the following C++ source file `s.cc`:
393
394 ```cpp
395 struct s {
396 int m;
397 int n;
398 };
399 int s::* p;
400 ```
401
402 When compiled with GCC and inspected with dwarfdump:
403
404 ```
405 $ g++ -gdwarf-5 -O3 -c s.cc
406 $ dwarfdump s.o
407 < 1><0x0000001e> DW_TAG_structure_type
408 DW_AT_name s
409 DW_AT_byte_size 0x00000008
410 DW_AT_sibling <0x0000003c>
411 < 2><0x00000029> DW_TAG_member
412 DW_AT_name m
413 DW_AT_type <0x0000003c>
414 DW_AT_data_member_location 0
415 < 2><0x00000032> DW_TAG_member
416 DW_AT_name n
417 DW_AT_type <0x0000003c>
418 DW_AT_data_member_location 4
419 < 1><0x0000003c> DW_TAG_base_type
420 DW_AT_byte_size 0x00000004
421 DW_AT_encoding DW_ATE_signed
422 DW_AT_name int
423 < 1><0x00000043> DW_TAG_ptr_to_member_type
424 DW_AT_containing_type <0x0000001e>
425 DW_AT_type <0x0000003c>
426 DW_AT_use_location len 0x0001: 22: DW_OP_plus
427 < 1><0x0000004e> DW_TAG_variable
428 DW_AT_name p
429 DW_AT_type <0x00000043>
430 DW_AT_external yes(1)
431 DW_AT_location len 0x0009: 030000000000000000: DW_OP_addr 0x00000000
432 ```
433
434 Note the location description for `DW_AT_use_location` is `DW_OP_plus`, which
435 reflects the GCC implementation of the pointer-to-member as an integral byte
436 offset within the containing object. For example, the value of `&s::m` in this
437 implementation is `offsetof(s, m)` and the value of `&s::n` is `offsetof(s, n)`:
438
439 ```cpp
440 struct s {
441 int m; // offsetof(s, m) == 0
442 int n; // offsetof(s, n) == 4
443 } o; // &o == 0xff00
444 int s::* p;
445 int *i;
446
447 p = &s::m; // p == 0
448 i = &(o.*p); // i == 0xff00 + 0
449 p = &s::n; // p == 4
450 i = &(o.*p); // i == 0xff00 + 4
451 ```
452
453 The expression `DW_OP_plus` accurately describes this implementation so long as
454 the entire containing object resides in memory in the default address space.
455
456 However, what if the containing object or the member pointed to are not at any
457 default address space address?
458
459 The compiler may store the containing object in memory in any address space,
460 in a register, recompute its value at each use, or compose any of these in
461 arbitrary ways.
462
463 The richness of the existing DWARF 5 expression language is a reflection of the
464 diversity of possible implementation strategies and optimization choices
465 affecting the location of an object in a program, and (modulo address spaces)
466 it can describe all of these locations for variables. However, the moment we
467 look at a pointer-to-member use we are restricted to only objects residing in a
468 contiguous piece of memory in the default address space.
469
470 To demonstrate the problem, consider a program which GCC chooses to optimize in
471 such a way that the containing object is not in memory at all:
472
473 ptm.h:
474 ```cpp
475 struct s {
476 int m;
477 int n;
478 };
479 void i(int);
480 extern int t;
481 void f(s x, int s::* p);
482 ```
483
484 ptm.cc:
485 ```cpp
486 #include "ptm.h"
487 void f(s x, int s::* p) {
488 for (int a = 0; a < t; ++a) {
489 x.m += a + x.n;
490 i(x.*p);
491 }
492 }
493 ```
494
495 main.cc:
496 ```cpp
497 #include "ptm.h"
498 int t = 100;
499 void i(int) {}
500 int main(int argc, char *argv[]) {
501 s x = { 0, 1 };
502 f(x, &s::m);
503 }
504 ```
505
506 When compiled and run under GDB:
507
508 ```
509 $ g++-9 -gdwarf-4 -O3 -c main.cc -o main.o
510 $ g++-9 -gdwarf-4 -O3 -c ptm.cc -o ptm.o
511 $ g++-9 main.o ptm.o -o use_location.out
512 $ gdb ./use_location.out
513 (gdb) maint set dwarf always-disassemble
514 (gdb) b ptm.cc:5
515 Breakpoint 1 at 0x119e: file ptm.cc, line 5.
516 (gdb) r
517
518 Breakpoint 1, f (x=..., p=<optimized out>) at ptm.cc:5
519 5 i(x.*p);
520 ```
521
522 Note that the compiler has promoted the entire object `x` into register `rdi`
523 for the body of the loop:
524
525 ```
526 (gdb) info addr x
527 Symbol "x" is multi-location:
528 Range 0x555555555160-0x5555555551af: a complex DWARF expression:
529 0: DW_OP_reg5 [$rdi]
530
531 Range 0x5555555551af-0x5555555551ba: a complex DWARF expression:
532 0: DW_OP_fbreg -56
533
534 .
535 (gdb) p $pc
536 $1 = (void (*)(void)) 0x55555555519e <f(s, int s::*)+62>
537 ```
538
539 And so it is impossible to interpret `DW_OP_use_location` in this case:
540
541 ```
542 (gdb) p x.*p
543 Address requested for identifier "x" which is in register $rdi
544 ```
545
546 With location descriptions on the stack, the definition of `DW_OP_use_location`
547 can be modified by replacing every instance of "address" with "location
548 description", as is described in [Type Entries](#type-entries).
549
550 To implement the fully generalized version of this attribute, GCC would only
551 need to change the expression from `DW_OP_plus` to `DW_OP_swap,
552 DW_OP_LLVM_offset`.
553
554 ### 3.2.7 Virtual Base Class
555
556 > NOTE: Without loss of generality, DWARF 4 is used in this example as full
557 > support for DWARF 5 is not present in the versions of the tools used. No
558 > feature of DWARF 5 provides a remedy for this issue.
559
560 This example highlights the inability of DWARF 5 to describe C++
561 virtual inheritance semantics.
562
563 The mechanism DWARF 5 provides for describing the location of an inherited
564 subobject is `DW_AT_data_member_location`. This attribute is overloaded to
565 describe both data member locations and inherited subobject locations, and
566 in each case has multiple possible forms:
567
568 * If an integral constant form, it encodes the byte offset from the derived
569 object to the data member or subobject.
570 * Otherwise, it encodes a location description to compute the address of the
571 data member or subobject given the address of the derived object.
572
573 Only the attribute describing a subobject, and only the location description
574 form are considered here.
575
576 In this case, when a debug agent wishes to locate the subobject, it first
577 pushes the address of the derived object onto the DWARF expression stack. It
578 then evaluates the location description associated with the
579 `DW_AT_data_member_location` of the `DW_TAG_inheritence` DIE corresponding to
580 the inherited subobject.
581
582 Consider the following C++ source file `ab.cc`:
583
584 ```cpp
585 class A {
586 public:
587 char x;
588 };
589 class B
590 : public virtual A {} o;
591 ```
592
593 When compiled with GCC and inspected with dwarfdump:
594
595 ```
596 $ g++ -gdwarf-5 -O3 -c ab.cc
597 $ dwarfdump ab.o
598 < 1><0x0000002a> DW_TAG_class_type
599 DW_AT_name A
600 DW_AT_byte_size 0x00000001
601 DW_AT_sibling <0x00000042>
602 < 1><0x00000049> DW_TAG_class_type
603 DW_AT_name B
604 DW_AT_byte_size 0x00000010
605 DW_AT_containing_type <0x00000049>
606 DW_AT_sibling <0x000000f9>
607 < 2><0x00000058> DW_TAG_inheritance
608 DW_AT_type <0x0000002a>
609 DW_AT_data_member_location len 0x0006: 1206481c0622:
610 DW_OP_dup DW_OP_deref DW_OP_lit24 DW_OP_minus DW_OP_deref DW_OP_plus
611 DW_AT_virtuality DW_VIRTUALITY_virtual
612 DW_AT_accessibility DW_ACCESS_public
613 ```
614
615 This `DW_AT_data_member_location` expression describes the dynamic process of
616 locating the `A`-in-`B` subobject according to the [Itanium
617 ABI](https://refspecs.linuxfoundation.org/cxxabi-1.86.html#layout). A diagram
618 of the logical layout of class `B` is:
619
620 ```
621 0: class B
622 0: vptr B
623 8: class A
624 8: A::x
625 ```
626
627 That is, the address of an object of class `B` is equivalent to the address for
628 the `vtable` pointer for `B`. As there are no other direct data members of `B`
629 the primary base class subobject of class `A` comes next, and there is no
630 intervening padding as the subobject alignment requirements are already
631 satisfied.
632
633 The `vtable` pointer for `B` contains an entry `vbase_offset` for each virtual
634 base class. In this case, that table layout is:
635
636 ```
637 -24: vbase_offset[A]=8
638 -16: offset_to_top=0
639 -8: B RTTI
640 0: <vtable for B>
641 ```
642
643 That is, to find the `vbase_offset` for the `A`-in-`B` subobject the address
644 `vptr B` is offset by the statically determined value `-24`.
645
646 Thus, in order to implement `DW_AT_data_member_location` for `A`-in-`B`, the
647 expression needs to index to a statically known byte offset of `-24` through
648 `vptr B` to lookup `vbase_offset` for `A`-in-`B`. It must then offset the
649 location of `B` by the dynamic value of `vbase_offset` (in this case `8`) to
650 arrive at the location of the inherited subobject.
651
652 This definition shares the same problem as example [3.2.6](#pointer-to-member)
653 in that it relies on the address of the derived object and inherited subobject,
654 when there is no guarantee either or both have any address at all.
655
656 To demonstrate the problem, consider a program which GCC chooses to optimize in
657 such a way that the derived object is not in memory at all:
658
659 f.h:
660 ```cpp
661 class A {
662 public:
663 char x;
664 };
665 class B
666 : public virtual A {};
667 void f(B b);
668 ```
669
670 f.cc:
671 ```cpp
672 #include "f.h"
673 void f(B b) {}
674 ```
675
676 main.cc:
677 ```cpp
678 #include "f.h"
679 int main(int argc, char *argv[]) {
680 B b;
681 b.x = 42;
682 f(b);
683 return b.x;
684 }
685 ```
686
687 When compiled and run under GDB:
688
689 ```
690 $ g++-9 -gdwarf-4 -O3 -c main.cc -o main.o
691 $ g++-9 -gdwarf-4 -O3 -c f.cc -o f.o
692 $ g++-9 main.o f.o -o cpp-vbase.out
693 (gdb) maint set dwarf always-disassemble
694 (gdb) b main.cc:6
695 Breakpoint 1 at 0x1090: file main.cc, line 6.
696 (gdb) r
697
698 Breakpoint 1, main (argc=<optimized out>, argv=<optimized out>) at main.cc:6
699 6 return b.x;
700 ```
701
702 Note that the compiler has elided storage for the entire object `x` in the
703 body of `main()`:
704
705 ```
706 (gdb) info addr b
707 Symbol "b" is multi-location:
708 Range 0x555555555078-0x5555555550af: a complex DWARF expression:
709 0: DW_OP_piece 8 (bytes)
710 2: DW_OP_const1u 42
711 4: DW_OP_stack_value
712 5: DW_OP_piece 1 (bytes)
713 7: DW_OP_piece 7 (bytes)
714
715 .
716 (gdb) p $pc
717 $1 = (void (*)(void)) 0x555555555090 <main(int, char**)+48>
718 ```
719
720 And so it is impossible to interpret `DW_OP_data_member_location` in this case:
721
722 ```
723 (gdb) p b
724 $2 = {<A> = <invalid address>, _vptr.B = <optimized out>}
725 ```
726
727 > NOTE: The `vptr B` which should occupy the first 8 bytes of the object `b`
728 > are undefined in the DWARF, but could be described as an implicit value by
729 > the compiler. This change would be trivial and would directly expose the
730 > issue in DWARF 5 described here.
731
732 With location descriptions on the stack, the definition of
733 `DW_OP_data_member_location` can be modified by replacing every instance of
734 "address" with "location description", as is described in [A.5 Type Entries](#type-entries).
735
736 To implement the fully generalized version of this attribute, GCC would only
737 need to change the last operation in the expression from `DW_OP_plus` to
738 `DW_OP_LLVM_offset`.
739
740 ## 3.3 Limitations
741
742 DWARF 5 is unable to describe variables in runtime indexed parts of registers.
743 This is required to describe a source variable that is located in a lane of a
744 SIMT vector register.
745
746 Some features only work when located in global memory. The type attribute
747 expressions require a base object which could be in any kind of location.
748
749 DWARF procedures can only accept global memory address arguments. This limits
750 the ability to factorize the creation of locations that involve other location
751 kinds.
752
753 There are no vector base types. This is required to describe vector registers.
754
755 There is no operation to create a memory location in a non-global address space.
756 Only the dereference operation supports providing an address space.
757
758 CFI location expressions do not allow composite locations or non-global address
759 space memory locations. Both these are needed in optimized code for devices with
760 vector registers and address spaces.
761
762 Bit field offsets are only supported in a limited way for register locations.
763 Supporting them in a uniform manner for all location kinds is required to
764 support languages with bit sized entities.
765
766 # 4. Extension Solution
767
768 This section outlines the extension to generalize the DWARF expression evaluation
769 model to allow location descriptions to be manipulated on the stack. It presents
770 a number of simplified examples to demonstrate the benefits and how the extension
771 solves the issues of heterogeneous devices. It presents how this is done in
772 a manner that is backwards compatible with DWARF 5.
773
774 ## 4.1 Location Description
775
776 In order to have consistent, composable operations that act on location
777 descriptions, the extension defines a uniform way to handle all location kinds.
778 That includes memory, register, implicit, implicit pointer, undefined, and
779 composite location descriptions.
780
781 Each kind of location description is conceptually a zero-based offset within a
782 piece of storage. The storage is a contiguous linear organization of a certain
783 number of bytes (see below for how this is extended to support bit sized
784 storage).
785
786 - For global memory, the storage is the linear stream of bytes of the
787 architecture's address size.
788 - For each separate architecture register, it is the linear stream of bytes of
789 the size of that specific register.
790 - For an implicit, it is the linear stream of bytes of the value when
791 represented using the value's base type which specifies the encoding, size,
792 and byte ordering.
793 - For undefined, it is an infinitely sized linear stream where every byte is
794 undefined.
795 - For composite, it is a linear stream of bytes defined by the composite's parts.
796
797 ## 4.2 Stack Location Description Operations
798
799 The DWARF expression stack is extended to allow each stack entry to either be a
800 value or a location description.
801
802 Evaluation rules are defined to implicitly convert a stack element that is a
803 value to a location description, or vice versa, so that all DWARF 5 expressions
804 continue to have the same semantics. This reflects that a memory address is
805 effectively used as a proxy for a memory location description.
806
807 For each place that allows a DWARF expression to be specified, it is defined if
808 the expression is to be evaluated as a value or a location description.
809
810 Existing DWARF expression operations that are used to act on memory addresses
811 are generalized to act on any location description kind. For example, the
812 DW_OP_deref operation pops a location description rather than a memory address
813 value from the stack and reads the storage associated with the location kind
814 starting at the location description's offset.
815
816 Existing DWARF expression operations that create location descriptions are
817 changed to pop and push location descriptions on the stack. For example, the
818 DW_OP_value, DW_OP_regx, DW_OP_implicit_value, DW_OP_implicit_pointer,
819 DW_OP_stack_value, and DW_OP_piece.
820
821 New operations that act on location descriptions can be added. For example, a
822 DW_OP_offset operation that modifies the offset of the location description on
823 top of the stack. Unlike the DW_OP_plus operation that only works with memory
824 address, a DW_OP_offset operation can work with any location kind.
825
826 To allow incremental and nested creation of composite location descriptions, a
827 DW_OP_piece_end can be defined to explicitly indicate the last part of a
828 composite. Currently, creating a composite must always be the last operation of
829 an expression.
830
831 A DW_OP_undefined operation can be defined that explicitly creates the undefined
832 location description. Currently this is only possible as a piece of a composite
833 when the stack is empty.
834
835 ## 4.3 Examples
836
837 This section provides some motivating examples to illustrate the benefits that
838 result from allowing location descriptions on the stack.
839
840 ### 4.3.1 Source Language Variable Spilled to Part of a Vector Register
841
842 A compiler generating code for a GPU may allocate a source language variable
843 that it proves has the same value for every lane of a SIMT thread in a scalar
844 register. It may then need to spill that scalar register. To avoid the high cost
845 of spilling to memory, it may spill to a fixed lane of one of the numerous
846 vector registers.
847
848 ![Source Language Variable Spilled to Part of a Vector Register Example](images/06-extension-spill-sgpr-to-static-vpgr-lane.example.png)
849
850 The following expression defines the location of a source language variable that
851 the compiler allocated in a scalar register, but had to spill to lane 5 of a
852 vector register at this point of the code.
853
854 DW_OP_regx VGPR0
855 DW_OP_offset_uconst 20
856
857 ![Source Language Variable Spilled to Part of a Vector Register Example: Step 1](images/06-extension-spill-sgpr-to-static-vpgr-lane.example.frame.1.png)
858
859 The DW_OP_regx pushes a register location description on the stack. The storage
860 for the register is the size of the vector register. The register location
861 description conceptually references that storage with an initial offset of 0.
862 The architecture defines the byte ordering of the register.
863
864 ![Source Language Variable Spilled to Part of a Vector Register Example: Step 2](images/06-extension-spill-sgpr-to-static-vpgr-lane.example.frame.2.png)
865
866 The DW_OP_offset_uconst pops a location description off the stack, adds its
867 operand value to the offset, and pushes the updated location description back on
868 the stack. In this case the source language variable is being spilled to lane 5
869 and each lane's component which is 32-bits (4 bytes), so the offset is 5*4=20.
870
871 ![Source Language Variable Spilled to Part of a Vector Register Example: Step 3](images/06-extension-spill-sgpr-to-static-vpgr-lane.example.frame.3.png)
872
873 The result of the expression evaluation is the location description on the top
874 of the stack.
875
876 An alternative approach could be for the target to define distinct register
877 names for each part of each vector register. However, this is not practical for
878 GPUs due to the sheer number of registers that would have to be defined. It
879 would also not permit a runtime index into part of the whole register to be used
880 as shown in the next example.
881
882 ### 4.3.2 Source Language Variable Spread Across Multiple Vector Registers
883
884 A compiler may generate SIMT code for a GPU. Each source language thread of
885 execution is mapped to a single lane of the GPU thread. Source language
886 variables that are mapped to a register, are mapped to the lane component of the
887 vector registers corresponding to the source language's thread of execution.
888
889 The location expression for such variables must therefore be executed in the
890 context of the focused source language thread of execution. A DW_OP_push_lane
891 operation can be defined to push the value of the lane for the currently focused
892 source language thread of execution. The value to use would be provided by the
893 consumer of DWARF when it evaluates the location expression.
894
895 If the source language variable is larger than the size of the vector register
896 lane component, then multiple vector registers are used. Each source language
897 thread of execution will only use the vector register components for its
898 associated lane.
899
900 ![Source Language Variable Spread Across Multiple Vector Registers Example](images/07-extension-multi-lane-vgpr.example.png)
901
902 The following expression defines the location of a source language variable that
903 has to occupy two vector registers. A composite location description is created
904 that combines the two parts. It will give the correct result regardless of which
905 lane corresponds to the source language thread of execution that the user is
906 focused on.
907
908 DW_OP_regx VGPR0
909 DW_OP_push_lane
910 DW_OP_uconst 4
911 DW_OP_mul
912 DW_OP_offset
913 DW_OP_piece 4
914 DW_OP_regx VGPR1
915 DW_OP_push_lane
916 DW_OP_uconst 4
917 DW_OP_mul
918 DW_OP_offset
919 DW_OP_piece 4
920
921 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 1](images/07-extension-multi-lane-vgpr.example.frame.1.png)
922
923 The DW_OP_regx VGPR0 pushes a location description for the first register.
924
925 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 2](images/07-extension-multi-lane-vgpr.example.frame.2.png)
926
927 The DW_OP_push_lane; DW_OP_uconst 4; DW_OP_mul calculates the offset for the
928 focused lanes vector register component as 4 times the lane number.
929
930 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 3](images/07-extension-multi-lane-vgpr.example.frame.3.png)
931
932 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 4](images/07-extension-multi-lane-vgpr.example.frame.4.png)
933
934 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 5](images/07-extension-multi-lane-vgpr.example.frame.5.png)
935
936 The DW_OP_offset adjusts the register location description's offset to the
937 runtime computed value.
938
939 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 6](images/07-extension-multi-lane-vgpr.example.frame.6.png)
940
941 The DW_OP_piece either creates a new composite location description, or adds a
942 new part to an existing incomplete one. It pops the location description to use
943 for the new part. It then pops the next stack element if it is an incomplete
944 composite location description, otherwise it creates a new incomplete composite
945 location description with no parts. Finally it pushes the incomplete composite
946 after adding the new part.
947
948 In this case a register location description is added to a new incomplete
949 composite location description. The 4 of the DW_OP_piece specifies the size of
950 the register storage that comprises the part. Note that the 4 bytes start at the
951 computed register offset.
952
953 For backwards compatibility, if the stack is empty or the top stack element is
954 an incomplete composite, an undefined location description is used for the part.
955 If the top stack element is a generic base type value, then it is implicitly
956 converted to a global memory location description with an offset equal to the
957 value.
958
959 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 7](images/07-extension-multi-lane-vgpr.example.frame.7.png)
960
961 The rest of the expression does the same for VGPR1. However, when the
962 DW_OP_piece is evaluated there is an incomplete composite on the stack. So the
963 VGPR1 register location description is added as a second part.
964
965 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 8](images/07-extension-multi-lane-vgpr.example.frame.8.png)
966
967 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 9](images/07-extension-multi-lane-vgpr.example.frame.9.png)
968
969 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 10](images/07-extension-multi-lane-vgpr.example.frame.10.png)
970
971 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 11](images/07-extension-multi-lane-vgpr.example.frame.11.png)
972
973 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 12](images/07-extension-multi-lane-vgpr.example.frame.12.png)
974
975 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 13](images/07-extension-multi-lane-vgpr.example.frame.13.png)
976
977 At the end of the expression, if the top stack element is an incomplete
978 composite location description, it is converted to a complete location
979 description and returned as the result.
980
981 ![Source Language Variable Spread Across Multiple Vector Registers Example: Step 14](images/07-extension-multi-lane-vgpr.example.frame.14.png)
982
983 ### 4.3.3 Source Language Variable Spread Across Multiple Kinds of Locations
984
985 This example is the same as the previous one, except the first 2 bytes of the
986 second vector register have been spilled to memory, and the last 2 bytes have
987 been proven to be a constant and optimized away.
988
989 ![Source Language Variable Spread Across Multiple Kinds of Locations Example](images/08-extension-mixed-composite.example.png)
990
991 DW_OP_regx VGPR0
992 DW_OP_push_lane
993 DW_OP_uconst 4
994 DW_OP_mul
995 DW_OP_offset
996 DW_OP_piece 4
997 DW_OP_addr 0xbeef
998 DW_OP_piece 2
999 DW_OP_uconst 0xf00d
1000 DW_OP_stack_value
1001 DW_OP_piece 2
1002 DW_OP_piece_end
1003
1004 The first 6 operations are the same.
1005
1006 ![Source Language Variable Spread Across Multiple Kinds of Locations Example: Step 7](images/08-extension-mixed-composite.example.frame.1.png)
1007
1008 The DW_OP_addr operation pushes a global memory location description on the
1009 stack with an offset equal to the address.
1010
1011 ![Source Language Variable Spread Across Multiple Kinds of Locations Example: Step 8](images/08-extension-mixed-composite.example.frame.2.png)
1012
1013 The next DW_OP_piece adds the global memory location description as the next 2
1014 byte part of the composite.
1015
1016 ![Source Language Variable Spread Across Multiple Kinds of Locations Example: Step 9](images/08-extension-mixed-composite.example.frame.3.png)
1017
1018 The DW_OP_uconst 0xf00d; DW_OP_stack_value pushes an implicit location
1019 description on the stack. The storage of the implicit location description is
1020 the representation of the value 0xf00d using the generic base type's encoding,
1021 size, and byte ordering.
1022
1023 ![Source Language Variable Spread Across Multiple Kinds of Locations Example: Step 10](images/08-extension-mixed-composite.example.frame.4.png)
1024
1025 ![Source Language Variable Spread Across Multiple Kinds of Locations Example: Step 11](images/08-extension-mixed-composite.example.frame.5.png)
1026
1027 The final DW_OP_piece adds 2 bytes of the implicit location description as the
1028 third part of the composite location description.
1029
1030 ![Source Language Variable Spread Across Multiple Kinds of Locations Example: Step 12](images/08-extension-mixed-composite.example.frame.6.png)
1031
1032 The DW_OP_piece_end operation explicitly makes the incomplete composite location
1033 description into a complete location description. This allows a complete
1034 composite location description to be created on the stack that can be used as
1035 the location description of another following operation. For example, the
1036 DW_OP_offset can be applied to it. More practically, it permits creation of
1037 multiple composite location descriptions on the stack which can be used to pass
1038 arguments to a DWARF procedure using a DW_OP_call* operation. This can be
1039 beneficial to factor the incrementally creation of location descriptions.
1040
1041 ![Source Language Variable Spread Across Multiple Kinds of Locations Example: Step 12](images/08-extension-mixed-composite.example.frame.7.png)
1042
1043 ### 4.3.4 Address Spaces
1044
1045 Heterogeneous devices can have multiple hardware supported address spaces which
1046 use specific hardware instructions to access them.
1047
1048 For example, GPUs that use SIMT execution may provide hardware support to access
1049 memory such that each lane can see a linear memory view, while the backing
1050 memory is actually being accessed in an interleaved manner so that the locations
1051 for each lanes Nth dword are contiguous. This minimizes cache lines read by the
1052 SIMT execution.
1053
1054 ![Address Spaces Example](images/09-extension-form-aspace.example.png)
1055
1056 The following expression defines the location of a source language variable that
1057 is allocated at offset 0x10 in the current subprograms stack frame. The
1058 subprogram stack frames are per lane and reside in an interleaved address space.
1059
1060 DW_OP_regval_type SGPR0 Generic
1061 DW_OP_uconst 1
1062 DW_OP_form_aspace_address
1063 DW_OP_offset 0x10
1064
1065 ![Address Spaces Example: Step 1](images/09-extension-form-aspace.example.frame.1.png)
1066
1067 The DW_OP_regval_type operation pushes the contents of SGPR0 as a generic value.
1068 This is the register that holds the address of the current stack frame.
1069
1070 ![Address Spaces Example: Step 2](images/09-extension-form-aspace.example.frame.2.png)
1071
1072 The DW_OP_uconst operation pushes the address space number. Each architecture
1073 defines the numbers it uses in DWARF. In this case, address space 1 is being
1074 used as the per lane memory.
1075
1076 ![Address Spaces Example: Step 3](images/09-extension-form-aspace.example.frame.3.png)
1077
1078 The DW_OP_form_aspace_address operation pops a value and an address space
1079 number. Each address space is associated with a separate storage. A memory
1080 location description is pushed which refers to the address space's storage, with
1081 an offset of the popped value.
1082
1083 ![Address Spaces Example: Step 4](images/09-extension-form-aspace.example.frame.4.png)
1084
1085 All operations that act on location descriptions work with memory locations
1086 regardless of their address space.
1087
1088 Every architecture defines address space 0 as the default global memory address
1089 space.
1090
1091 Generalizing memory location descriptions to include an address space component
1092 avoids having to create specialized operations to work with address spaces.
1093
1094 The source variable is at offset 0x10 in the stack frame. The DW_OP_offset
1095 operation works on memory location descriptions that have an address space just
1096 like for any other kind of location description.
1097
1098 ![Address Spaces Example: Step 5](images/09-extension-form-aspace.example.frame.5.png)
1099
1100 The only operations in DWARF 5 that take an address space are DW_OP_xderef*.
1101 They treat a value as the address in a specified address space, and read its
1102 contents. There is no operation to actually create a location description that
1103 references an address space. There is no way to include address space memory
1104 locations in parts of composite locations.
1105
1106 Since DW_OP_piece now takes any kind of location description for its pieces, it
1107 is now possible for parts of a composite to involve locations in different
1108 address spaces. For example, this can happen when parts of a source variable
1109 allocated in a register are spilled to a stack frame that resides in the
1110 non-global address space.
1111
1112 ### 4.3.5 Bit Offsets
1113
1114 With the generalization of location descriptions on the stack, it is possible to
1115 define a DW_OP_bit_offset operation that adjusts the offset of any kind of
1116 location in terms of bits rather than bytes. The offset can be a runtime
1117 computed value. This is generally useful for any source language that support
1118 bit sized entities, and for registers that are not a whole number of bytes.
1119
1120 DWARF 5 only supports bit fields in composites using DW_OP_bit_piece. It does
1121 not support runtime computed offsets which can happen for bit field packed
1122 arrays. It is also not generally composable as it must be the last part of an
1123 expression.
1124
1125 The following example defines a location description for a source variable that
1126 is allocated starting at bit 20 of a register. A similar expression could be
1127 used if the source variable was at a bit offset within memory or a particular
1128 address space, or if the offset is a runtime value.
1129
1130 ![Bit Offsets Example](images/10-extension-bit-offset.example.png)
1131
1132 DW_OP_regx SGPR3
1133 DW_OP_uconst 20
1134 DW_OP_bit_offset
1135
1136 ![Bit Offsets Example: Step 1](images/10-extension-bit-offset.example.frame.1.png)
1137
1138 ![Bit Offsets Example: Step 2](images/10-extension-bit-offset.example.frame.2.png)
1139
1140 ![Bit Offsets Example: Step 3](images/10-extension-bit-offset.example.frame.3.png)
1141
1142 The DW_OP_bit_offset operation pops a value and location description from the
1143 stack. It pushes the location description after updating its offset using the
1144 value as a bit count.
1145
1146 ![Bit Offsets Example: Step 4](images/10-extension-bit-offset.example.frame.4.png)
1147
1148 The ordering of bits within a byte, like byte ordering, is defined by the target
1149 architecture. A base type could be extended to specify bit ordering in addition
1150 to byte ordering.
1151
1152 ## 4.4 Call Frame Information (CFI)
1153
1154 DWARF defines call frame information (CFI) that can be used to virtually unwind
1155 the subprogram call stack. This involves determining the location where register
1156 values have been spilled. DWARF 5 limits these locations to either be registers
1157 or global memory. As shown in the earlier examples, heterogeneous devices may
1158 spill registers to parts of other registers, to non-global memory address
1159 spaces, or even a composite of different location kinds.
1160
1161 Therefore, the extension extends the CFI rules to support any kind of location
1162 description, and operations to create locations in address spaces.
1163
1164 ## 4.5 Objects Not In Byte Aligned Global Memory
1165
1166 DWARF 5 only effectively supports byte aligned memory locations on the stack by
1167 using a global memory address as a proxy for a memory location description. This
1168 is a problem for attributes that define DWARF expressions that require the
1169 location of some source language entity that is not allocated in byte aligned
1170 global memory.
1171
1172 For example, the DWARF expression of the DW_AT_data_member_location attribute is
1173 evaluated with an initial stack containing the location of a type instance
1174 object. That object could be located in a register, in a non-global memory
1175 address space, be described by a composite location description, or could even
1176 be an implicit location description.
1177
1178 A similar problem exists for DWARF expressions that use the
1179 DW_OP_push_object_address operation. This operation pushes the location of a
1180 program object associated with the attribute that defines the expression.
1181
1182 Allowing any kind of location description on the stack permits the DW_OP_call*
1183 operations to be used to factor the creation of location descriptions. The
1184 inputs and outputs of the call are passed on the stack. For example, on GPUs an
1185 expression can be defined to describe the effective PC of inactive lanes of SIMT
1186 execution. This is naturally done by composing the result of expressions for
1187 each nested control flow region. This can be done by making each control flow
1188 region have its own DWARF procedure, and then calling it from the expressions of
1189 the nested control flow regions. The alternative is to make each control flow
1190 region have the complete expression which results in much larger DWARF and is
1191 less convenient to generate.
1192
1193 GPU compilers work hard to allocate objects in the larger number of registers to
1194 reduce memory accesses, they have to use different memory address spaces, and
1195 they perform optimizations that result in composites of these. Allowing
1196 operations to work with any kind of location description enables creating
1197 expressions that support all of these.
1198
1199 Full general support for bit fields and implicit locations benefits
1200 optimizations on any target.
1201
1202 ## 4.6 Higher Order Operations
1203
1204 The generalization allows an elegant way to add higher order operations that
1205 create location descriptions out of other location descriptions in a general
1206 composable manner.
1207
1208 For example, a DW_OP_extend operation could create a composite location
1209 description out of a location description, an element size, and an element
1210 count. The resulting composite would effectively be a vector of element count
1211 elements with each element being the same location description of the specified
1212 bit size.
1213
1214 A DW_OP_select_bit_piece operation could create a composite location description
1215 out of two location descriptions, a bit mask value, and an element size. The
1216 resulting composite would effectively be a vector of elements, selecting from
1217 one of the two input locations according to the bit mask.
1218
1219 These could be used in the expression of an attribute that computes the
1220 effective PC of lanes of SIMT execution. The vector result efficiently computes
1221 the PC for each SIMT lane at once. The mask could be the hardware execution mask
1222 register that controls which SIMT lanes are executing. For active divergent
1223 lanes the vector element would be the current PC, and for inactive divergent
1224 lanes the PC would correspond to the source language line at which the lane is
1225 logically positioned.
1226
1227 Similarly, a DW_OP_overlay_piece operation could be defined that creates a
1228 composite location description out of two location descriptions, an offset
1229 value, and a size. The resulting composite would consist of parts that are
1230 equivalent to one of the location descriptions, but with the other location
1231 description replacing a slice defined by the offset and size. This could be used
1232 to efficiently express a source language array that has had a set of elements
1233 promoted into a vector register when executing a set of iterations of a loop in
1234 a SIMD manner.
1235
1236 ## 4.7 Objects In Multiple Places
1237
1238 A compiler may allocate a source variable in stack frame memory, but for some
1239 range of code may promote it to a register. If the generated code does not
1240 change the register value, then there is no need to save it back to memory.
1241 Effectively, during that range, the source variable is in both memory and a
1242 register. If a consumer, such as a debugger, allows the user to change the value
1243 of the source variable in that PC range, then it would need to change both
1244 places.
1245
1246 DWARF 5 supports loclists which are able to specify the location of a source
1247 language entity is in different places at different PC locations. It can also
1248 express that a source language entity is in multiple places at the same time.
1249
1250 DWARF 5 defines operation expressions and loclists separately. In general, this
1251 is adequate as non-memory location descriptions can only be computed as the last
1252 step of an expression evaluation.
1253
1254 However, allowing location descriptions on the stack permits non-memory location
1255 descriptions to be used in the middle of expression evaluation. For example, the
1256 DW_OP_call* and DW_OP_implicit_pointer operations can result in evaluating the
1257 expression of a DW_AT_location attribute of a DIE. The DW_AT_location attribute
1258 allows the loclist form. So the result could include multiple location
1259 descriptions.
1260
1261 Similarly, the DWARF expression associated with attributes such as
1262 DW_AT_data_member_location that are evaluated with an initial stack containing a
1263 location description, or a DWARF operation expression that uses the
1264 DW_OP_push_object_address operation, may want to act on the result of another
1265 expression that returned a location description involving multiple places.
1266
1267 Therefore, the extension needs to define how expression operations that use those
1268 results will behave. The extension does this by generalizing the expression stack
1269 to allow an entry to be one or more single location descriptions. In doing this,
1270 it unifies the definitions of DWARF operation expressions and loclist
1271 expressions in a natural way.
1272
1273 All operations that act on location descriptions are extended to act on multiple
1274 single location descriptions. For example, the DW_OP_offset operation adds the
1275 offset to each single location description. The DW_OP_deref* operations simply
1276 read the storage of one of the single location descriptions, since multiple
1277 single location descriptions must all hold the same value. Similarly, if the
1278 evaluation of a DWARF expression results in multiple single location
1279 descriptions, the consumer can ensure any updates are done to all of them, and
1280 any reads can use any one of them.
1281
1282 # 5. Conclusion
1283
1284 A strength of DWARF is that it has generally sought to provide generalized
1285 composable solutions that address many problems, rather than solutions that only
1286 address one-off issues. This extension attempts to follow that tradition by
1287 defining a backwards compatible composable generalization that can address a
1288 significant family of issues. It addresses the specific issues present for
1289 heterogeneous computing devices, provides benefits for non-heterogeneous
1290 devices, and can help address a number of other previously reported issues.
1291
1292 # A. Changes to DWARF Debugging Information Format Version 5
1293
1294 > NOTE: This appendix provides changes relative to DWARF Version 5. It has been
1295 > defined such that it is backwards compatible with DWARF Version 5.
1296 > Non-normative text is shown in <i>italics</i>. The section numbers generally
1297 > correspond to those in the DWARF Version 5 standard unless specified
1298 > otherwise. Definitions are given to clarify how existing expression
1299 > operations, CFI operations, and attributes behave with respect to generalized
1300 > location descriptions that support multiple places.
1301 >
1302 > > NOTE: Notes are included to describe how the changes are to be applied to
1303 > > the DWARF Version 5 standard. They also describe rational and issues that
1304 > > may need further consideration.
1305
1306 ## A.2 General Description
1307
1308 ### A.2.5 DWARF Expressions
1309
1310 > NOTE: This section, and its nested sections, replaces DWARF Version 5 section
1311 > 2.5 and section 2.6. It is based on the text of the existing DWARF Version 5
1312 > standard.
1313
1314 DWARF expressions describe how to compute a value or specify a location.
1315
1316 <i>The evaluation of a DWARF expression can provide the location of an object,
1317 the value of an array bound, the length of a dynamic string, the desired value
1318 itself, and so on.</i>
1319
1320 If the evaluation of a DWARF expression does not encounter an error, then it can
1321 either result in a value (see [2.5.2 DWARF Expression
1322 Value](#dwarf-expression-value)) or a location description (see [2.5.3 DWARF
1323 Location Description](#dwarf-location-description)). When a DWARF expression
1324 is evaluated, it may be specified whether a value or location description is
1325 required as the result kind.
1326
1327 If a result kind is specified, and the result of the evaluation does not match
1328 the specified result kind, then the implicit conversions described in
1329 [2.5.4.4.3 Memory Location Description Operations](#memory-location-description-operations)
1330 are performed if valid. Otherwise, the DWARF expression is ill-formed.
1331
1332 If the evaluation of a DWARF expression encounters an evaluation error, then the
1333 result is an evaluation error.
1334
1335 > NOTE: Decided to define the concept of an evaluation error. An alternative is
1336 > to introduce an undefined value base type in a similar way to location
1337 > descriptions having an undefined location description. Then operations that
1338 > encounter an evaluation error can return the undefined location description or
1339 > value with an undefined base type.
1340 >
1341 > All operations that act on values would return an undefined entity if given an
1342 > undefined value. The expression would then always evaluate to completion, and
1343 > can be tested to determine if it is an undefined entity.
1344 >
1345 > However, this would add considerable additional complexity and does not match
1346 > that GDB throws an exception when these evaluation errors occur.
1347
1348 If a DWARF expression is ill-formed, then the result is undefined.
1349
1350 The following sections detail the rules for when a DWARF expression is
1351 ill-formed or results in an evaluation error.
1352
1353 A DWARF expression can either be encoded as an operation expression (see [2.5.4
1354 DWARF Operation Expressions](#dwarf-operation-expressions)), or as a
1355 location list expression (see [2.5.5 DWARF Location List
1356 Expressions](#dwarf-location-list-expressions)).
1357
1358 #### A.2.5.1 DWARF Expression Evaluation Context
1359
1360 A DWARF expression is evaluated in a context that can include a number of
1361 context elements. If multiple context elements are specified then they must be
1362 self consistent or the result of the evaluation is undefined. The context
1363 elements that can be specified are:
1364
1365 1. <i>A current result kind</i>
1366
1367 The kind of result required by the DWARF expression evaluation. If specified
1368 it can be a location description or a value.
1369
1370 2. <i>A current thread</i>
1371
1372 The target architecture thread identifier of the source program thread of
1373 execution for which a user presented expression is currently being
1374 evaluated.
1375
1376 It is required for operations that are related to target architecture
1377 threads.
1378
1379 <i>For example, the `DW_OP_regval_type` operation.</i>
1380
1381 3. <i>A current call frame</i>
1382
1383 The target architecture call frame identifier. It identifies a call frame
1384 that corresponds to an active invocation of a subprogram in the current
1385 thread. It is identified by its address on the call stack. The address is
1386 referred to as the Canonical Frame Address (CFA). The call frame information
1387 is used to determine the CFA for the call frames of the current thread's
1388 call stack (see [6.4 Call Frame Information](#call-frame-information)).
1389
1390 It is required for operations that specify target architecture registers to
1391 support virtual unwinding of the call stack.
1392
1393 <i>For example, the `DW_OP_*reg*` operations.</i>
1394
1395 If specified, it must be an active call frame in the current thread.
1396 Otherwise the result is undefined.
1397
1398 If it is the currently executing call frame, then it is termed the top call
1399 frame.
1400
1401 4. <i>A current program location</i>
1402
1403 The target architecture program location corresponding to the current call
1404 frame of the current thread.
1405
1406 The program location of the top call frame is the target architecture
1407 program counter for the current thread. The call frame information is used
1408 to obtain the value of the return address register to determine the program
1409 location of the other call frames (see [6.4 Call Frame
1410 Information](#call-frame-information)).
1411
1412 It is required for the evaluation of location list expressions to select
1413 amongst multiple program location ranges. It is required for operations that
1414 specify target architecture registers to support virtual unwinding of the
1415 call stack (see [6.4 Call Frame Information](#call-frame-information)).
1416
1417 If specified:
1418
1419 - If the current call frame is the top call frame, it must be the current
1420 target architecture program location.
1421 - If the current call frame F is not the top call frame, it must be the
1422 program location associated with the call site in the current caller frame
1423 F that invoked the callee frame.
1424 - Otherwise the result is undefined.
1425
1426 5. <i>A current compilation unit</i>
1427
1428 The compilation unit debug information entry that contains the DWARF
1429 expression being evaluated.
1430
1431 It is required for operations that reference debug information associated
1432 with the same compilation unit, including indicating if such references use
1433 the 32-bit or 64-bit DWARF format. It can also provide the default address
1434 space address size if no current target architecture is specified.
1435
1436 <i>For example, the `DW_OP_constx` and `DW_OP_addrx` operations.</i>
1437
1438 <i>Note that this compilation unit may not be the same as the compilation
1439 unit determined from the loaded code object corresponding to the current
1440 program location. For example, the evaluation of the expression E associated
1441 with a `DW_AT_location` attribute of the debug information entry operand of
1442 the `DW_OP_call*` operations is evaluated with the compilation unit that
1443 contains E and not the one that contains the `DW_OP_call*` operation
1444 expression.</i>
1445
1446 6. <i>A current target architecture</i>
1447
1448 The target architecture.
1449
1450 It is required for operations that specify target architecture specific
1451 entities.
1452
1453 <i>For example, target architecture specific entities include DWARF register
1454 identifiers, DWARF address space identifiers, the default address space, and
1455 the address space address sizes.</i>
1456
1457 If specified:
1458
1459 - If the current frame is specified, then the current target architecture
1460 must be the same as the target architecture of the current frame.
1461
1462 - If the current frame is specified and is the top frame, and if the current
1463 thread is specified, then the current target architecture must be the same
1464 as the target architecture of the current thread.
1465
1466 - If the current compilation unit is specified, then the current target
1467 architecture default address space address size must be the same as the
1468 `address_size` field in the header of the current compilation unit and any
1469 associated entry in the `.debug_aranges` section.
1470 - If the current program location is specified, then the current target
1471 architecture must be the same as the target architecture of any line
1472 number information entry (see [6.2 Line Number
1473 Information](#line-number-information)) corresponding to the current
1474 program location.
1475 - If the current program location is specified, then the current target
1476 architecture default address space address size must be the same as the
1477 `address_size` field in the header of any entry corresponding to the
1478 current program location in the `.debug_addr`, `.debug_line`,
1479 `.debug_rnglists`, `.debug_rnglists.dwo`, `.debug_loclists`, and
1480 `.debug_loclists.dwo` sections.
1481 - Otherwise the result is undefined.
1482
1483 7. <i>A current object</i>
1484
1485 The location description of a program object.
1486
1487 It is required for the `DW_OP_push_object_address` operation.
1488
1489 <i>For example, the `DW_AT_data_location` attribute on type debug
1490 information entries specifies the program object corresponding to a runtime
1491 descriptor as the current object when it evaluates its associated
1492 expression.</i>
1493
1494 The result is undefined if the location description is invalid (see [2.5.3
1495 DWARF Location Description](#dwarf-location-description)).
1496
1497 8. <i>An initial stack</i>
1498
1499 This is a list of values or location descriptions that will be pushed on the
1500 operation expression evaluation stack in the order provided before
1501 evaluation of an operation expression starts.
1502
1503 Some debugger information entries have attributes that evaluate their DWARF
1504 expression value with initial stack entries. In all other cases the initial
1505 stack is empty.
1506
1507 The result is undefined if any location descriptions are invalid (see [2.5.3
1508 DWARF Location Description](#dwarf-location-description)).
1509
1510 If the evaluation requires a context element that is not specified, then the
1511 result of the evaluation is an error.
1512
1513 <i>A DWARF expression for a location description may be able to be evaluated
1514 without a thread, call frame, program location, or architecture context. For
1515 example, the location of a global variable may be able to be evaluated without
1516 such context. If the expression evaluates with an error then it may indicate the
1517 variable has been optimized and so requires more context.</i>
1518
1519 <i>The DWARF expression for call frame information (see [6.4 Call Frame
1520 Information](#call-frame-information)) operations are restricted to those
1521 that do not require the compilation unit context to be specified.</i>
1522
1523 The DWARF is ill-formed if all the `address_size` fields in the headers of all
1524 the entries in the `.debug_info`, `.debug_addr`, `.debug_line`,
1525 `.debug_rnglists`, `.debug_rnglists.dwo`, `.debug_loclists`, and
1526 `.debug_loclists.dwo` sections corresponding to any given program location do
1527 not match.
1528
1529 #### A.2.5.2 DWARF Expression Value
1530
1531 A value has a type and a literal value. It can represent a literal value of any
1532 supported base type of the target architecture. The base type specifies the
1533 size, encoding, and endianity of the literal value.
1534
1535 > NOTE: It may be desirable to add an implicit pointer base type encoding. It
1536 > would be used for the type of the value that is produced when the
1537 > `DW_OP_deref*` operation retrieves the full contents of an implicit pointer
1538 > location storage created by the `DW_OP_implicit_pointer` operation. The
1539 > literal value would record the debugging information entry and byte
1540 > displacement specified by the associated `DW_OP_implicit_pointer` operation.
1541
1542 There is a distinguished base type termed the generic type, which is an integral
1543 type that has the size of an address in the target architecture default address
1544 space, a target architecture defined endianity, and unspecified signedness.
1545
1546 <i>The generic type is the same as the unspecified type used for stack
1547 operations defined in DWARF Version 4 and before.</i>
1548
1549 An integral type is a base type that has an encoding of `DW_ATE_signed`,
1550 `DW_ATE_signed_char`, `DW_ATE_unsigned`, `DW_ATE_unsigned_char`,
1551 `DW_ATE_boolean`, or any target architecture defined integral encoding in the
1552 inclusive range `DW_ATE_lo_user` to `DW_ATE_hi_user`.
1553
1554 > NOTE: It is unclear if `DW_ATE_address` is an integral type. GDB does not seem
1555 > to consider it as integral.
1556
1557 #### A.2.5.3 DWARF Location Description
1558
1559 <i>Debugging information must provide consumers a way to find the location of
1560 program variables, determine the bounds of dynamic arrays and strings, and
1561 possibly to find the base address of a subprogram's call frame or the return
1562 address of a subprogram. Furthermore, to meet the needs of recent computer
1563 architectures and optimization techniques, debugging information must be able to
1564 describe the location of an object whose location changes over the object's
1565 lifetime, and may reside at multiple locations simultaneously during parts of an
1566 object's lifetime.</i>
1567
1568 Information about the location of program objects is provided by location
1569 descriptions.
1570
1571 Location descriptions can consist of one or more single location descriptions.
1572
1573 A single location description specifies the location storage that holds a
1574 program object and a position within the location storage where the program
1575 object starts. The position within the location storage is expressed as a bit
1576 offset relative to the start of the location storage.
1577
1578 A location storage is a linear stream of bits that can hold values. Each
1579 location storage has a size in bits and can be accessed using a zero-based bit
1580 offset. The ordering of bits within a location storage uses the bit numbering
1581 and direction conventions that are appropriate to the current language on the
1582 target architecture.
1583
1584 There are five kinds of location storage:
1585
1586 1. <i>memory location storage</i>
1587
1588 Corresponds to the target architecture memory address spaces.
1589
1590 2. <i>register location storage</i>
1591
1592 Corresponds to the target architecture registers.
1593
1594 3. <i>implicit location storage</i>
1595
1596 Corresponds to fixed values that can only be read.
1597
1598 4. <i>undefined location storage</i>
1599
1600 Indicates no value is available and therefore cannot be read or written.
1601
1602 5. <i>composite location storage</i>
1603
1604 Allows a mixture of these where some bits come from one location storage and
1605 some from another location storage, or from disjoint parts of the same
1606 location storage.
1607
1608 > NOTE: It may be better to add an implicit pointer location storage kind used
1609 > by the `DW_OP_implicit_pointer` operation. It would specify the debugger
1610 > information entry and byte offset provided by the operations.
1611
1612 <i>Location descriptions are a language independent representation of addressing
1613 rules.</i>
1614
1615 - <i>They can be the result of evaluating a debugger information entry attribute
1616 that specifies an operation expression of arbitrary complexity. In this usage
1617 they can describe the location of an object as long as its lifetime is either
1618 static or the same as the lexical block (see [3.5 Lexical Block
1619 Entries](#lexical-block-entries)) that owns it, and it does not move during
1620 its lifetime.</i>
1621
1622 - <i>They can be the result of evaluating a debugger information entry attribute
1623 that specifies a location list expression. In this usage they can describe the
1624 location of an object that has a limited lifetime, changes its location during
1625 its lifetime, or has multiple locations over part or all of its lifetime.</i>
1626
1627 If a location description has more than one single location description, the
1628 DWARF expression is ill-formed if the object value held in each single location
1629 description's position within the associated location storage is not the same
1630 value, except for the parts of the value that are uninitialized.
1631
1632 <i>A location description that has more than one single location description can
1633 only be created by a location list expression that has overlapping program
1634 location ranges, or certain expression operations that act on a location
1635 description that has more than one single location description. There are no
1636 operation expression operations that can directly create a location description
1637 with more than one single location description.</i>
1638
1639 <i>A location description with more than one single location description can be
1640 used to describe objects that reside in more than one piece of storage at the
1641 same time. An object may have more than one location as a result of
1642 optimization. For example, a value that is only read may be promoted from memory
1643 to a register for some region of code, but later code may revert to reading the
1644 value from memory as the register may be used for other purposes. For the code
1645 region where the value is in a register, any change to the object value must be
1646 made in both the register and the memory so both regions of code will read the
1647 updated value.</i>
1648
1649 <i>A consumer of a location description with more than one single location
1650 description can read the object's value from any of the single location
1651 descriptions (since they all refer to location storage that has the same value),
1652 but must write any changed value to all the single location descriptions.</i>
1653
1654 Updating a location description L by a bit offset B is defined as adding the
1655 value of B to the bit offset of each single location description SL of L. It is
1656 an evaluation error if the updated bit offset of any SL is less than 0 or
1657 greater than or equal to the size of the location storage specified by SL.
1658
1659 The evaluation of an expression may require context elements to create a
1660 location description. If such a location description is accessed, the storage it
1661 denotes is that associated with the context element values specified when the
1662 location description was created, which may differ from the context at the time
1663 it is accessed.
1664
1665 <i>For example, creating a register location description requires the thread
1666 context: the location storage is for the specified register of that thread.
1667 Creating a memory location description for an address space may required a
1668 thread context: the location storage is the memory associated with that
1669 thread.</i>
1670
1671 If any of the context elements required to create a location description change,
1672 the location description becomes invalid and accessing it is undefined.
1673
1674 <i>Examples of context that can invalidate a location description are:</i>
1675
1676 - <i>The thread context is required and execution causes the thread to
1677 terminate.</i>
1678 - <i>The call frame context is required and further execution causes the call
1679 frame to return to the calling frame.</i>
1680 - <i>The program location is required and further execution of the thread
1681 occurs. That could change the location list entry or call frame information
1682 entry that applies.</i>
1683 - <i>An operation uses call frame information:</i>
1684 - <i>Any of the frames used in the virtual call frame unwinding return.</i>
1685 - <i>The top call frame is used, the program location is used to select the
1686 call frame information entry, and further execution of the thread
1687 occurs.</i>
1688
1689 <i>A DWARF expression can be used to compute a location description for an
1690 object. A subsequent DWARF expression evaluation can be given the object
1691 location description as the object context or initial stack context to compute a
1692 component of the object. The final result is undefined if the object location
1693 description becomes invalid between the two expression evaluations.</i>
1694
1695 A change of a thread's program location may not make a location description
1696 invalid, yet may still render it as no longer meaningful. Accessing such a
1697 location description, or using it as the object context or initial stack context
1698 of an expression evaluation, may produce an undefined result.
1699
1700 <i>For example, a location description may specify a register that no longer
1701 holds the intended program object after a program location change. One way to
1702 avoid such problems is to recompute location descriptions associated with
1703 threads when their program locations change.</i>
1704
1705 #### A.2.5.4 DWARF Operation Expressions
1706
1707 An operation expression is comprised of a stream of operations, each consisting
1708 of an opcode followed by zero or more operands. The number of operands is
1709 implied by the opcode.
1710
1711 Operations represent a postfix operation on a simple stack machine. Each stack
1712 entry can hold either a value or a location description. Operations can act on
1713 entries on the stack, including adding entries and removing entries. If the kind
1714 of a stack entry does not match the kind required by the operation and is not
1715 implicitly convertible to the required kind
1716 (see [2.5.4.4.3 Memory Location Description Operations](#memory-location-description-operations)),
1717 then the DWARF operation expression is ill-formed.
1718
1719 Evaluation of an operation expression starts with an empty stack on which the
1720 entries from the initial stack provided by the context are pushed in the order
1721 provided. Then the operations are evaluated, starting with the first operation
1722 of the stream. Evaluation continues until either an operation has an evaluation
1723 error, or until one past the last operation of the stream is reached.
1724
1725 The result of the evaluation is:
1726
1727 - If an operation has an evaluation error, or an operation evaluates an
1728 expression that has an evaluation error, then the result is an evaluation
1729 error.
1730 - If the current result kind specifies a location description, then:
1731 - If the stack is empty, the result is a location description with one
1732 undefined location description.
1733
1734 <i>This rule is for backwards compatibility with DWARF Version 5 which uses
1735 an empty operation expression for this purpose.</i>
1736
1737 - If the top stack entry is a location description, or can be converted to one
1738 (see [2.5.4.4.3 Memory Location Description Operations](#memory-location-description-operations)),
1739 then the result is that, possibly converted, location description. Any other entries on the
1740 stack are discarded.
1741 - Otherwise the DWARF expression is ill-formed.
1742
1743 > NOTE: Could define this case as returning an implicit location description
1744 > as if the `DW_OP_implicit` operation is performed.
1745
1746 - If the current result kind specifies a value, then:
1747 - If the top stack entry is a value, or can be converted to one (see
1748 [2.5.4.4.3 Memory Location Description Operations](#memory-location-description-operations)),
1749 then the result is that, possibly converted, value. Any other entries on the stack are
1750 discarded.
1751 - Otherwise the DWARF expression is ill-formed.
1752 - If the current result kind is not specified, then:
1753 - If the stack is empty, the result is a location description with one
1754 undefined location description.
1755
1756 <i>This rule is for backwards compatibility with DWARF Version 5 which uses
1757 an empty operation expression for this purpose.</i>
1758
1759 > NOTE: This rule is consistent with the rule above for when a location
1760 > description is requested. However, GDB appears to report this as an error
1761 > and no GDB tests appear to cause an empty stack for this case.
1762
1763 - Otherwise, the top stack entry is returned. Any other entries on the stack
1764 are discarded.
1765
1766 An operation expression is encoded as a byte block with some form of prefix that
1767 specifies the byte count. It can be used:
1768
1769 - as the value of a debugging information entry attribute that is encoded using
1770 class `exprloc` (see [7.5.5 Classes and Forms](#classes-and-forms)),
1771 - as the operand to certain operation expression operations,
1772 - as the operand to certain call frame information operations (see [6.4 Call
1773 Frame Information](#call-frame-information)),
1774 - and in location list entries (see [2.5.5 DWARF Location List
1775 Expressions](#dwarf-location-list-expressions)).
1776
1777 ##### A.2.5.4.1 Stack Operations
1778
1779 > NOTE: This section replaces DWARF Version 5 section 2.5.1.3.
1780
1781 The following operations manipulate the DWARF stack. Operations that index the
1782 stack assume that the top of the stack (most recently added entry) has index 0.
1783 They allow the stack entries to be either a value or location description.
1784
1785 If any stack entry accessed by a stack operation is an incomplete composite
1786 location description (see [2.5.4.4.6 Composite Location Description Operations]
1787 (#composite-location-description-operations)), then the DWARF expression is ill-formed.
1788
1789 > NOTE: These operations now support stack entries that are values and location
1790 > descriptions.
1791
1792 > NOTE: If it is desired to also make them work with incomplete composite
1793 > location descriptions, then would need to define that the composite location
1794 > storage specified by the incomplete composite location description is also
1795 > replicated when a copy is pushed. This ensures that each copy of the
1796 > incomplete composite location description can update the composite location
1797 > storage they specify independently.
1798
1799 1. `DW_OP_dup`
1800
1801 `DW_OP_dup` duplicates the stack entry at the top of the stack.
1802
1803 2. `DW_OP_drop`
1804
1805 `DW_OP_drop` pops the stack entry at the top of the stack and discards it.
1806
1807 3. `DW_OP_pick`
1808
1809 `DW_OP_pick` has a single unsigned 1-byte operand that represents an index
1810 I. A copy of the stack entry with index I is pushed onto the stack.
1811
1812 4. `DW_OP_over`
1813
1814 `DW_OP_over` pushes a copy of the entry with index 1.
1815
1816 <i>This is equivalent to a `DW_OP_pick 1` operation.</i>
1817
1818 5. `DW_OP_swap`
1819
1820 `DW_OP_swap` swaps the top two stack entries. The entry at the top of the
1821 stack becomes the second stack entry, and the second stack entry becomes the
1822 top of the stack.
1823
1824 6. `DW_OP_rot`
1825
1826 `DW_OP_rot` rotates the first three stack entries. The entry at the top of
1827 the stack becomes the third stack entry, the second entry becomes the top of
1828 the stack, and the third entry becomes the second entry.
1829
1830 <i>Examples illustrating many of these stack operations are found in Appendix
1831 D.1.2 on page 289.</i>
1832
1833 ##### A.2.5.4.2 Control Flow Operations
1834
1835 > NOTE: This section replaces DWARF Version 5 section 2.5.1.5.
1836
1837 The following operations provide simple control of the flow of a DWARF operation
1838 expression.
1839
1840 1. `DW_OP_nop`
1841
1842 `DW_OP_nop` is a place holder. It has no effect on the DWARF stack entries.
1843
1844 2. `DW_OP_le`, `DW_OP_ge`, `DW_OP_eq`, `DW_OP_lt`, `DW_OP_gt`,
1845 `DW_OP_ne`
1846
1847 > NOTE: The same as in DWARF Version 5 section 2.5.1.5.
1848
1849 3. `DW_OP_skip`
1850
1851 `DW_OP_skip` is an unconditional branch. Its single operand is a 2-byte
1852 signed integer constant. The 2-byte constant is the number of bytes of the
1853 DWARF expression to skip forward or backward from the current operation,
1854 beginning after the 2-byte constant.
1855
1856 If the updated position is at one past the end of the last operation, then
1857 the operation expression evaluation is complete.
1858
1859 Otherwise, the DWARF expression is ill-formed if the updated operation
1860 position is not in the range of the first to last operation inclusive, or
1861 not at the start of an operation.
1862
1863 4. `DW_OP_bra`
1864
1865 `DW_OP_bra` is a conditional branch. Its single operand is a 2-byte signed
1866 integer constant. This operation pops the top of stack. If the value popped
1867 is not the constant 0, the 2-byte constant operand is the number of bytes of
1868 the DWARF operation expression to skip forward or backward from the current
1869 operation, beginning after the 2-byte constant.
1870
1871 If the updated position is at one past the end of the last operation, then
1872 the operation expression evaluation is complete.
1873
1874 Otherwise, the DWARF expression is ill-formed if the updated operation
1875 position is not in the range of the first to last operation inclusive, or
1876 not at the start of an operation.
1877
1878 5. `DW_OP_call2, DW_OP_call4, DW_OP_call_ref`
1879
1880 `DW_OP_call2`, `DW_OP_call4`, and `DW_OP_call_ref` perform DWARF procedure
1881 calls during evaluation of a DWARF operation expression.
1882
1883 `DW_OP_call2` and `DW_OP_call4`, have one operand that is, respectively, a
1884 2-byte or 4-byte unsigned offset DR that represents the byte offset of a
1885 debugging information entry D relative to the beginning of the current
1886 compilation unit.
1887
1888 `DW_OP_call_ref` has one operand that is a 4-byte unsigned value in the
1889 32-bit DWARF format, or an 8-byte unsigned value in the 64-bit DWARF format,
1890 that represents the byte offset DR of a debugging information entry D
1891 relative to the beginning of the `.debug_info` section that contains the
1892 current compilation unit. D may not be in the current compilation unit.
1893
1894 > NOTE: DWARF Version 5 states that DR can be an offset in a `.debug_info`
1895 > section other than the one that contains the current compilation unit. It
1896 > states that relocation of references from one executable or shared object
1897 > file to another must be performed by the consumer. But given that DR is
1898 > defined as an offset in a `.debug_info` section this seems impossible. If
1899 > DR was defined as an implementation defined value, then the consumer could
1900 > choose to interpret the value in an implementation defined manner to
1901 > reference a debug information in another executable or shared object.
1902 >
1903 > In ELF the `.debug_info` section is in a non-`PT_LOAD` segment so standard
1904 > dynamic relocations cannot be used. But even if they were loaded segments
1905 > and dynamic relocations were used, DR would need to be the address of D,
1906 > not an offset in a `.debug_info` section. That would also need DR to be
1907 > the size of a global address. So it would not be possible to use the
1908 > 32-bit DWARF format in a 64-bit global address space. In addition, the
1909 > consumer would need to determine what executable or shared object the
1910 > relocated address was in so it could determine the containing compilation
1911 > unit.
1912 >
1913 > GDB only interprets DR as an offset in the `.debug_info` section that
1914 > contains the current compilation unit.
1915 >
1916 > This comment also applies to `DW_OP_implicit_pointer`.
1917
1918 <i>Operand interpretation of `DW_OP_call2`, `DW_OP_call4`, and
1919 `DW_OP_call_ref` is exactly like that for `DW_FORM_ref2`, `DW_FORM_ref4`,
1920 and `DW_FORM_ref_addr`, respectively.</i>
1921
1922 The call operation is evaluated by:
1923
1924 - If D has a `DW_AT_location` attribute that is encoded as a `exprloc` that
1925 specifies an operation expression E, then execution of the current
1926 operation expression continues from the first operation of E. Execution
1927 continues until one past the last operation of E is reached, at which
1928 point execution continues with the operation following the call operation.
1929 The operations of E are evaluated with the same current context, except
1930 current compilation unit is the one that contains D and the stack is the
1931 same as that being used by the call operation. After the call operation
1932 has been evaluated, the stack is therefore as it is left by the evaluation
1933 of the operations of E. Since E is evaluated on the same stack as the call
1934 operation, E can use, and/or remove entries already on the stack, and can
1935 add new entries to the stack.
1936
1937 <i>Values on the stack at the time of the call may be used as parameters
1938 by the called expression and values left on the stack by the called
1939 expression may be used as return values by prior agreement between the
1940 calling and called expressions.</i>
1941
1942 - If D has a `DW_AT_location` attribute that is encoded as a `loclist` or
1943 `loclistsptr`, then the specified location list expression E is evaluated.
1944 The evaluation of E uses the current context, except the result kind is a
1945 location description, the compilation unit is the one that contains D, and
1946 the initial stack is empty. The location description result is pushed on
1947 the stack.
1948
1949 > NOTE: This rule avoids having to define how to execute a matched
1950 > location list entry operation expression on the same stack as the call
1951 > when there are multiple matches. But it allows the call to obtain the
1952 > location description for a variable or formal parameter which may use a
1953 > location list expression.
1954 >
1955 > An alternative is to treat the case when D has a `DW_AT_location`
1956 > attribute that is encoded as a `loclist` or `loclistsptr`, and the
1957 > specified location list expression E' matches a single location list
1958 > entry with operation expression E, the same as the `exprloc` case and
1959 > evaluate on the same stack.
1960 >
1961 > But this is not attractive as if the attribute is for a variable that
1962 > happens to end with a non-singleton stack, it will not simply put a
1963 > location description on the stack. Presumably the intent of using
1964 > `DW_OP_call*` on a variable or formal parameter debugger information
1965 > entry is to push just one location description on the stack. That
1966 > location description may have more than one single location description.
1967 >
1968 > The previous rule for `exprloc` also has the same problem, as normally a
1969 > variable or formal parameter location expression may leave multiple
1970 > entries on the stack and only return the top entry.
1971 >
1972 > GDB implements `DW_OP_call*` by always executing E on the same stack. If
1973 > the location list has multiple matching entries, it simply picks the
1974 > first one and ignores the rest. This seems fundamentally at odds with
1975 > the desire to support multiple places for variables.
1976 >
1977 > So, it feels like `DW_OP_call*` should both support pushing a location
1978 > description on the stack for a variable or formal parameter, and also
1979 > support being able to execute an operation expression on the same stack.
1980 > Being able to specify a different operation expression for different
1981 > program locations seems a desirable feature to retain.
1982 >
1983 > A solution to that is to have a distinct `DW_AT_proc` attribute for the
1984 > `DW_TAG_dwarf_procedure` debugging information entry. Then the
1985 > `DW_AT_location` attribute expression is always executed separately and
1986 > pushes a location description (that may have multiple single location
1987 > descriptions), and the `DW_AT_proc` attribute expression is always
1988 > executed on the same stack and can leave anything on the stack.
1989 >
1990 > The `DW_AT_proc` attribute could have the new classes `exprproc`,
1991 > `loclistproc`, and `loclistsptrproc` to indicate that the expression is
1992 > executed on the same stack. `exprproc` is the same encoding as
1993 > `exprloc`. `loclistproc` and `loclistsptrproc` are the same encoding as
1994 > their non-`proc` counterparts, except the DWARF is ill-formed if the
1995 > location list does not match exactly one location list entry and a
1996 > default entry is required. These forms indicate explicitly that the
1997 > matched single operation expression must be executed on the same stack.
1998 > This is better than ad hoc special rules for `loclistproc` and
1999 > `loclistsptrproc` which are currently clearly defined to always return a
2000 > location description. The producer then explicitly indicates the intent
2001 > through the attribute classes.
2002 >
2003 > Such a change would be a breaking change for how GDB implements
2004 > `DW_OP_call*`. However, are the breaking cases actually occurring in
2005 > practice? GDB could implement the current approach for DWARF Version 5,
2006 > and the new semantics for DWARF Version 6 which has been done for some
2007 > other features.
2008 >
2009 > Another option is to limit the execution to be on the same stack only to
2010 > the evaluation of an expression E that is the value of a
2011 > `DW_AT_location` attribute of a `DW_TAG_dwarf_procedure` debugging
2012 > information entry. The DWARF would be ill-formed if E is a location list
2013 > expression that does not match exactly one location list entry. In all
2014 > other cases the evaluation of an expression E that is the value of a
2015 > `DW_AT_location` attribute would evaluate E with the current context,
2016 > except the result kind is a location description, the compilation unit
2017 > is the one that contains D, and the initial stack is empty. The location
2018 > description result is pushed on the stack.
2019
2020 - If D has a `DW_AT_const_value` attribute with a value V, then it is as if
2021 a `DW_OP_implicit_value V` operation was executed.
2022
2023 <i>This allows a call operation to be used to compute the location
2024 description for any variable or formal parameter regardless of whether the
2025 producer has optimized it to a constant. This is consistent with the
2026 `DW_OP_implicit_pointer` operation.</i>
2027
2028 > NOTE: Alternatively, could deprecate using `DW_AT_const_value` for
2029 > `DW_TAG_variable` and `DW_TAG_formal_parameter` debugger information
2030 > entries that are constants and instead use `DW_AT_location` with an
2031 > operation expression that results in a location description with one
2032 > implicit location description. Then this rule would not be required.
2033
2034 - Otherwise, there is no effect and no changes are made to the stack.
2035
2036 > NOTE: In DWARF Version 5, if D does not have a `DW_AT_location` then
2037 > `DW_OP_call*` is defined to have no effect. It is unclear that this is
2038 > the right definition as a producer should be able to rely on using
2039 > `DW_OP_call*` to get a location description for any
2040 > non-`DW_TAG_dwarf_procedure` debugging information entries. Also, the
2041 > producer should not be creating DWARF with `DW_OP_call*` to a
2042 > `DW_TAG_dwarf_procedure` that does not have a `DW_AT_location`
2043 > attribute. So, should this case be defined as an ill-formed DWARF
2044 > expression?
2045
2046 <i>The `DW_TAG_dwarf_procedure` debugging information entry can be used to
2047 define DWARF procedures that can be called.</i>
2048
2049 ##### A.2.5.4.3 Value Operations
2050
2051 This section describes the operations that push values on the stack.
2052
2053 Each value stack entry has a type and a literal value. It can represent a
2054 literal value of any supported base type of the target architecture. The base
2055 type specifies the size, encoding, and endianity of the literal value.
2056
2057 The base type of value stack entries can be the distinguished generic type.
2058
2059 ###### A.2.5.4.3.1 Literal Operations
2060
2061 > NOTE: This section replaces DWARF Version 5 section 2.5.1.1.
2062
2063 The following operations all push a literal value onto the DWARF stack.
2064
2065 Operations other than `DW_OP_const_type` push a value V with the generic type.
2066 If V is larger than the generic type, then V is truncated to the generic type
2067 size and the low-order bits used.
2068
2069 1. `DW_OP_lit0`, `DW_OP_lit1`, ..., `DW_OP_lit31`
2070
2071 `DW_OP_lit<N>` operations encode an unsigned literal value N from 0 through
2072 31, inclusive. They push the value N with the generic type.
2073
2074 2. `DW_OP_const1u`, `DW_OP_const2u`, `DW_OP_const4u`, `DW_OP_const8u`
2075
2076 `DW_OP_const<N>u` operations have a single operand that is a 1, 2, 4, or
2077 8-byte unsigned integer constant U, respectively. They push the value U with
2078 the generic type.
2079
2080 3. `DW_OP_const1s`, `DW_OP_const2s`, `DW_OP_const4s`, `DW_OP_const8s`
2081
2082 `DW_OP_const<N>s` operations have a single operand that is a 1, 2, 4, or
2083 8-byte signed integer constant S, respectively. They push the value S with
2084 the generic type.
2085
2086 4. `DW_OP_constu`
2087
2088 `DW_OP_constu` has a single unsigned LEB128 integer operand N. It pushes the
2089 value N with the generic type.
2090
2091 5. `DW_OP_consts`
2092
2093 `DW_OP_consts` has a single signed LEB128 integer operand N. It pushes the
2094 value N with the generic type.
2095
2096 6. `DW_OP_constx`
2097
2098 `DW_OP_constx` has a single unsigned LEB128 integer operand that represents
2099 a zero-based index into the `.debug_addr` section relative to the value of
2100 the `DW_AT_addr_base` attribute of the associated compilation unit. The
2101 value N in the `.debug_addr` section has the size of the generic type. It
2102 pushes the value N with the generic type.
2103
2104 <i>The `DW_OP_constx` operation is provided for constants that require
2105 link-time relocation but should not be interpreted by the consumer as a
2106 relocatable address (for example, offsets to thread-local storage).</i>
2107
2108 7. `DW_OP_const_type`
2109
2110 `DW_OP_const_type` has three operands. The first is an unsigned LEB128
2111 integer DR that represents the byte offset of a debugging information entry
2112 D relative to the beginning of the current compilation unit, that provides
2113 the type T of the constant value. The second is a 1-byte unsigned integral
2114 constant S. The third is a block of bytes B, with a length equal to S.
2115
2116 TS is the bit size of the type T. The least significant TS bits of B are
2117 interpreted as a value V of the type D. It pushes the value V with the type
2118 D.
2119
2120 The DWARF is ill-formed if D is not a `DW_TAG_base_type` debugging
2121 information entry in the current compilation unit, or if TS divided by 8
2122 (the byte size) and rounded up to a whole number is not equal to S.
2123
2124 <i>While the size of the byte block B can be inferred from the type D
2125 definition, it is encoded explicitly into the operation so that the
2126 operation can be parsed easily without reference to the `.debug_info`
2127 section.</i>
2128
2129 ###### A.2.5.4.3.2 Arithmetic and Logical Operations
2130
2131 > NOTE: This section is the same as DWARF Version 5 section 2.5.1.4.
2132
2133 ###### A.2.5.4.3.3 Type Conversion Operations
2134
2135 > NOTE: This section is the same as DWARF Version 5 section 2.5.1.6.
2136
2137 ###### A.2.5.4.3.4 Special Value Operations
2138
2139 > NOTE: This section replaces parts of DWARF Version 5 sections 2.5.1.2,
2140 2.5.1.3, and 2.5.1.7.
2141
2142 There are these special value operations currently defined:
2143
2144 1. `DW_OP_regval_type`
2145
2146 `DW_OP_regval_type` has two operands. The first is an unsigned LEB128
2147 integer that represents a register number R. The second is an unsigned
2148 LEB128 integer DR that represents the byte offset of a debugging information
2149 entry D relative to the beginning of the current compilation unit, that
2150 provides the type T of the register value.
2151
2152 The operation is equivalent to performing `DW_OP_regx R; DW_OP_deref_type
2153 DR`.
2154
2155 > NOTE: Should DWARF allow the type T to be a larger size than the size of
2156 > the register R? Restricting a larger bit size avoids any issue of
2157 > conversion as the, possibly truncated, bit contents of the register is
2158 > simply interpreted as a value of T. If a conversion is wanted it can be
2159 > done explicitly using a `DW_OP_convert` operation.
2160 >
2161 > GDB has a per register hook that allows a target specific conversion on a
2162 > register by register basis. It defaults to truncation of bigger registers.
2163 > Removing use of the target hook does not cause any test failures in common
2164 > architectures. If the compiler for a target architecture did want some
2165 > form of conversion, including a larger result type, it could always
2166 > explicitly use the `DW_OP_convert` operation.
2167 >
2168 > If T is a larger type than the register size, then the default GDB
2169 > register hook reads bytes from the next register (or reads out of bounds
2170 > for the last register!). Removing use of the target hook does not cause
2171 > any test failures in common architectures (except an illegal hand written
2172 > assembly test). If a target architecture requires this behavior, these
2173 > extensions allow a composite location description to be used to combine
2174 > multiple registers.
2175
2176 2. `DW_OP_deref`
2177
2178 S is the bit size of the generic type divided by 8 (the byte size) and
2179 rounded up to a whole number. DR is the offset of a hypothetical debug
2180 information entry D in the current compilation unit for a base type of the
2181 generic type.
2182
2183 The operation is equivalent to performing `DW_OP_deref_type S, DR`.
2184
2185 3. `DW_OP_deref_size`
2186
2187 `DW_OP_deref_size` has a single 1-byte unsigned integral constant that
2188 represents a byte result size S.
2189
2190 TS is the smaller of the generic type bit size and S scaled by 8 (the byte
2191 size). If TS is smaller than the generic type bit size then T is an unsigned
2192 integral type of bit size TS, otherwise T is the generic type. DR is the
2193 offset of a hypothetical debug information entry D in the current
2194 compilation unit for a base type T.
2195
2196 > NOTE: Truncating the value when S is larger than the generic type matches
2197 > what GDB does. This allows the generic type size to not be an integral
2198 > byte size. It does allow S to be arbitrarily large. Should S be restricted
2199 > to the size of the generic type rounded up to a multiple of 8?
2200
2201 The operation is equivalent to performing `DW_OP_deref_type S, DR`, except
2202 if T is not the generic type, the value V pushed is zero-extended to the
2203 generic type bit size and its type changed to the generic type.
2204
2205 4. `DW_OP_deref_type`
2206
2207 `DW_OP_deref_type` has two operands. The first is a 1-byte unsigned integral
2208 constant S. The second is an unsigned LEB128 integer DR that represents the
2209 byte offset of a debugging information entry D relative to the beginning of
2210 the current compilation unit, that provides the type T of the result value.
2211
2212 TS is the bit size of the type T.
2213
2214 <i>While the size of the pushed value V can be inferred from the type T, it
2215 is encoded explicitly as the operand S so that the operation can be parsed
2216 easily without reference to the `.debug_info` section.</i>
2217
2218 > NOTE: It is unclear why the operand S is needed. Unlike
2219 > `DW_OP_const_type`, the size is not needed for parsing. Any evaluation
2220 > needs to get the base type T to push with the value to know its encoding
2221 > and bit size.
2222
2223 It pops one stack entry that must be a location description L.
2224
2225 A value V of TS bits is retrieved from the location storage LS specified by
2226 one of the single location descriptions SL of L.
2227
2228 <i>If L, or the location description of any composite location description
2229 part that is a subcomponent of L, has more than one single location
2230 description, then any one of them can be selected as they are required to
2231 all have the same value. For any single location description SL, bits are
2232 retrieved from the associated storage location starting at the bit offset
2233 specified by SL. For a composite location description, the retrieved bits
2234 are the concatenation of the N bits from each composite location part PL,
2235 where N is limited to the size of PL.</i>
2236
2237 V is pushed on the stack with the type T.
2238
2239 > NOTE: This definition makes it an evaluation error if L is a register
2240 > location description that has less than TS bits remaining in the register
2241 > storage. Particularly since these extensions extend location descriptions
2242 > to have a bit offset, it would be odd to define this as performing sign
2243 > extension based on the type, or be target architecture dependent, as the
2244 > number of remaining bits could be any number. This matches the GDB
2245 > implementation for `DW_OP_deref_type`.
2246 >
2247 > These extensions define `DW_OP_*breg*` in terms of `DW_OP_regval_type`.
2248 > `DW_OP_regval_type` is defined in terms of `DW_OP_regx`, which uses a 0
2249 > bit offset, and `DW_OP_deref_type`. Therefore, it requires the register
2250 > size to be greater or equal to the address size of the address space. This
2251 > matches the GDB implementation for `DW_OP_*breg*`.
2252
2253 The DWARF is ill-formed if D is not in the current compilation unit, D is
2254 not a `DW_TAG_base_type` debugging information entry, or if TS divided by 8
2255 (the byte size) and rounded up to a whole number is not equal to S.
2256
2257 > NOTE: This definition allows the base type to be a bit size since there
2258 > seems no reason to restrict it.
2259
2260 It is an evaluation error if any bit of the value is retrieved from the
2261 undefined location storage or the offset of any bit exceeds the size of the
2262 location storage LS specified by any single location description SL of L.
2263
2264 See [2.5.4.4.5 Implicit Location Description Operations](#implicit-location-description-operations)
2265 for special rules concerning implicit location descriptions created by the
2266 `DW_OP_implicit_pointer` operation.
2267
2268 5. `DW_OP_xderef`
2269
2270 `DW_OP_xderef` pops two stack entries. The first must be an integral type
2271 value that represents an address A. The second must be an integral type
2272 value that represents a target architecture specific address space
2273 identifier AS.
2274
2275 The address size S is defined as the address bit size of the target
2276 architecture specific address space that corresponds to AS.
2277
2278 A is adjusted to S bits by zero extending if necessary, and then treating
2279 the least significant S bits as an unsigned value A'.
2280
2281 It creates a location description L with one memory location description SL.
2282 SL specifies the memory location storage LS that corresponds to AS with a
2283 bit offset equal to A' scaled by 8 (the byte size).
2284
2285 If AS is an address space that is specific to context elements, then LS
2286 corresponds to the location storage associated with the current context.
2287
2288 <i>For example, if AS is for per thread storage then LS is the location
2289 storage for the current thread. Therefore, if L is accessed by an operation,
2290 the location storage selected when the location description was created is
2291 accessed, and not the location storage associated with the current context
2292 of the access operation.</i>
2293
2294 The DWARF expression is ill-formed if AS is not one of the values defined by
2295 the target architecture specific `DW_ASPACE_*` values.
2296
2297 The operation is equivalent to popping A and AS, pushing L, and then
2298 performing `DW_OP_deref`. The value V retrieved is left on the stack with
2299 the generic type.
2300
2301 6. `DW_OP_xderef_size`
2302
2303 `DW_OP_xderef_size` has a single 1-byte unsigned integral constant that
2304 represents a byte result size S.
2305
2306 It pops two stack entries. The first must be an integral type value
2307 that represents an address A. The second must be an integral type
2308 value that represents a target architecture specific address space
2309 identifier AS.
2310
2311 It creates a location description L as described for `DW_OP_xderef`.
2312
2313 The operation is equivalent to popping A and AS, pushing L, and then
2314 performing `DW_OP_deref_size S` . The zero-extended value V retrieved is
2315 left on the stack with the generic type.
2316
2317 7. `DW_OP_xderef_type`
2318
2319 `DW_OP_xderef_type` has two operands. The first is a 1-byte unsigned
2320 integral constant S. The second operand is an unsigned LEB128 integer DR
2321 that represents the byte offset of a debugging information entry D relative
2322 to the beginning of the current compilation unit, that provides the type T
2323 of the result value.
2324
2325 It pops two stack entries. The first must be an integral type value that
2326 represents an address A. The second must be an integral type value that
2327 represents a target architecture specific address space identifier AS.
2328
2329 It creates a location description L as described for `DW_OP_xderef`.
2330
2331 The operation is equivalent to popping A and AS, pushing L, and then
2332 performing `DW_OP_deref_type DR` . The value V retrieved is left on the
2333 stack with the type T.
2334
2335 8. `DW_OP_entry_value` <i>Deprecated</i>
2336
2337 `DW_OP_entry_value` pushes the value of an expression that is evaluated in
2338 the context of the calling frame.
2339
2340 <i>It may be used to determine the value of arguments on entry to the
2341 current call frame provided they are not clobbered.</i>
2342
2343 It has two operands. The first is an unsigned LEB128 integer S. The second
2344 is a block of bytes, with a length equal S, interpreted as a DWARF operation
2345 expression E.
2346
2347 E is evaluated with the current context, except the result kind is
2348 unspecified, the call frame is the one that called the current frame, the
2349 program location is the call site in the calling frame, the object is
2350 unspecified, and the initial stack is empty. The calling frame information
2351 is obtained by virtually unwinding the current call frame using the call
2352 frame information (see [6.4 Call Frame
2353 Information](#call-frame-information)).
2354
2355 If the result of E is a location description L (see [2.5.4.4.4 Register
2356 Location Description
2357 Operations](#register-location-description-operations)), and the last
2358 operation executed by E is a `DW_OP_reg*` for register R with a target
2359 architecture specific base type of T, then the contents of the register are
2360 retrieved as if a `DW_OP_deref_type DR` operation was performed where DR is
2361 the offset of a hypothetical debug information entry in the current
2362 compilation unit for T. The resulting value V is pushed on the stack.
2363
2364 <i>Using `DW_OP_reg*` provides a more compact form for the case where the
2365 value was in a register on entry to the subprogram.</i>
2366
2367 > NOTE: It is unclear how this provides a more compact expression, as
2368 > `DW_OP_regval_type` could be used which is marginally larger.
2369
2370 If the result of E is a value V, then V is pushed on the stack.
2371
2372 Otherwise, the DWARF expression is ill-formed.
2373
2374 <i>The `DW_OP_entry_value` operation is deprecated as its main usage is
2375 provided by other means. DWARF Version 5 added the
2376 `DW_TAG_call_site_parameter` debugger information entry for call sites that
2377 has `DW_AT_call_value`, `DW_AT_call_data_location`, and
2378 `DW_AT_call_data_value` attributes that provide DWARF expressions to compute
2379 actual parameter values at the time of the call, and requires the producer
2380 to ensure the expressions are valid to evaluate even when virtually
2381 unwound.</i>
2382
2383 > NOTE: GDB only implements `DW_OP_entry_value` when E is exactly
2384 > `DW_OP_reg*` or `DW_OP_breg*; DW_OP_deref*`.
2385
2386 ##### A.2.5.4.4 Location Description Operations
2387
2388 This section describes the operations that push location descriptions on the
2389 stack.
2390
2391 ###### A.2.5.4.4.1 General Location Description Operations
2392
2393 > NOTE: This section replaces part of DWARF Version 5 section 2.5.1.3.
2394
2395 1. `DW_OP_push_object_address`
2396
2397 `DW_OP_push_object_address` pushes the location description L of the current
2398 object.
2399
2400 <i>This object may correspond to an independent variable that is part of a
2401 user presented expression that is being evaluated. The object location
2402 description may be determined from the variable's own debugging information
2403 entry or it may be a component of an array, structure, or class whose
2404 address has been dynamically determined by an earlier step during user
2405 expression evaluation.</i>
2406
2407 <i>This operation provides explicit functionality (especially for arrays
2408 involving descriptors) that is analogous to the implicit push of the base
2409 location description of a structure prior to evaluation of a
2410 `DW_AT_data_member_location` to access a data member of a structure.</i>
2411
2412 > NOTE: This operation could be removed and the object location description
2413 > specified as the initial stack as for `DW_AT_data_member_location`.
2414 >
2415 > Or this operation could be used instead of needing to specify an initial
2416 > stack. The latter approach is more composable as access to the object may
2417 > be needed at any point of the expression, and passing it as the initial
2418 > stack requires the entire expression to be aware where on the stack it is.
2419 > If this were done, ``DW_AT_use_location`` would require a
2420 > ``DW_OP_push_object2_address`` operation for the second object.
2421 >
2422 > Or a more general way to pass an arbitrary number of arguments in and an
2423 > operation to get the Nth one such as ``DW_OP_arg N``. A vector of
2424 > arguments would then be passed in the expression context rather than an
2425 > initial stack. This could also resolve the issues with ``DW_OP_call*`` by
2426 > allowing a specific number of arguments passed in and returned to be
2427 > specified. The ``DW_OP_call*`` operation could then always execute on a
2428 > separate stack: the number of arguments would be specified in a new call
2429 > operation and taken from the callers stack, and similarly the number of
2430 > return results specified and copied from the called stack back to the
2431 > callee stack when the called expression was complete.
2432 >
2433 > The only attribute that specifies a current object is
2434 > `DW_AT_data_location` so the non-normative text seems to overstate how
2435 > this is being used. Or are there other attributes that need to state they
2436 > pass an object?
2437
2438 ###### A.2.5.4.4.2 Undefined Location Description Operations
2439
2440 > NOTE: This section replaces DWARF Version 5 section 2.6.1.1.1.
2441
2442 <i>The undefined location storage represents a piece or all of an object that is
2443 present in the source but not in the object code (perhaps due to optimization).
2444 Neither reading nor writing to the undefined location storage is meaningful.</i>
2445
2446 An undefined location description specifies the undefined location storage.
2447 There is no concept of the size of the undefined location storage, nor of a bit
2448 offset for an undefined location description. The `DW_OP_*piece` operations can
2449 implicitly specify an undefined location description, allowing any size and
2450 offset to be specified, and results in a part with all undefined bits.
2451
2452 ###### A.2.5.4.4.3 Memory Location Description Operations
2453
2454 > NOTE: This section replaces parts of DWARF Version 5 section 2.5.1.1, 2.5.1.2,
2455 > 2.5.1.3, and 2.6.1.1.2.
2456
2457 Each of the target architecture specific address spaces has a corresponding
2458 memory location storage that denotes the linear addressable memory of that
2459 address space. The size of each memory location storage corresponds to the range
2460 of the addresses in the corresponding address space.
2461
2462 <i>It is target architecture defined how address space location storage maps to
2463 target architecture physical memory. For example, they may be independent
2464 memory, or more than one location storage may alias the same physical memory
2465 possibly at different offsets and with different interleaving. The mapping may
2466 also be dictated by the source language address classes.</i>
2467
2468 A memory location description specifies a memory location storage. The bit
2469 offset corresponds to a bit position within a byte of the memory. Bits accessed
2470 using a memory location description, access the corresponding target
2471 architecture memory starting at the bit position within the byte specified by
2472 the bit offset.
2473
2474 A memory location description that has a bit offset that is a multiple of 8 (the
2475 byte size) is defined to be a byte address memory location description. It has a
2476 memory byte address A that is equal to the bit offset divided by 8.
2477
2478 A memory location description that does not have a bit offset that is a multiple
2479 of 8 (the byte size) is defined to be a bit field memory location description.
2480 It has a bit position B equal to the bit offset modulo 8, and a memory byte
2481 address A equal to the bit offset minus B that is then divided by 8.
2482
2483 The address space AS of a memory location description is defined to be the
2484 address space that corresponds to the memory location storage associated with
2485 the memory location description.
2486
2487 A location description that is comprised of one byte address memory location
2488 description SL is defined to be a memory byte address location description. It
2489 has a byte address equal to A and an address space equal to AS of the
2490 corresponding SL.
2491
2492 `DW_ASPACE_none` is defined as the target architecture default address space.
2493
2494 If a stack entry is required to be a location description, but it is a value V
2495 with the generic type, then it is implicitly converted to a location description
2496 L with one memory location description SL. SL specifies the memory location
2497 storage that corresponds to the target architecture default address space with a
2498 bit offset equal to V scaled by 8 (the byte size).
2499
2500 > NOTE: If it is wanted to allow any integral type value to be implicitly
2501 > converted to a memory location description in the target architecture default
2502 > address space:
2503 >
2504 > > If a stack entry is required to be a location description, but is a value V
2505 > > with an integral type, then it is implicitly converted to a location
2506 > > description L with a one memory location description SL. If the type size of
2507 > > V is less than the generic type size, then the value V is zero extended to
2508 > > the size of the generic type. The least significant generic type size bits
2509 > > are treated as an unsigned value to be used as an address A. SL specifies
2510 > > memory location storage corresponding to the target architecture default
2511 > > address space with a bit offset equal to A scaled by 8 (the byte size).
2512 >
2513 > The implicit conversion could also be defined as target architecture specific.
2514 > For example, GDB checks if V is an integral type. If it is not it gives an
2515 > error. Otherwise, GDB zero-extends V to 64 bits. If the GDB target defines a
2516 > hook function, then it is called. The target specific hook function can modify
2517 > the 64-bit value, possibly sign extending based on the original value type.
2518 > Finally, GDB treats the 64-bit value V as a memory location address.
2519
2520 If a stack entry is required to be a location description, but it is an implicit
2521 pointer value IPV with the target architecture default address space, then it is
2522 implicitly converted to a location description with one single location
2523 description specified by IPV. See
2524 [2.5.4.4.5 Implicit Location Description Operations](#implicit-location-description-operations).
2525
2526 If a stack entry is required to be a value, but it is a location description L
2527 with one memory location description SL in the target architecture default
2528 address space with a bit offset B that is a multiple of 8, then it is implicitly
2529 converted to a value equal to B divided by 8 (the byte size) with the generic
2530 type.
2531
2532 1. `DW_OP_addr`
2533
2534 `DW_OP_addr` has a single byte constant value operand, which has the size of
2535 the generic type, that represents an address A.
2536
2537 It pushes a location description L with one memory location description SL
2538 on the stack. SL specifies the memory location storage corresponding to the
2539 target architecture default address space with a bit offset equal to A
2540 scaled by 8 (the byte size).
2541
2542 <i>If the DWARF is part of a code object, then A may need to be relocated.
2543 For example, in the ELF code object format, A must be adjusted by the
2544 difference between the ELF segment virtual address and the virtual address
2545 at which the segment is loaded.</i>
2546
2547 2. `DW_OP_addrx`
2548
2549 `DW_OP_addrx` has a single unsigned LEB128 integer operand that represents a
2550 zero-based index into the `.debug_addr` section relative to the value of the
2551 `DW_AT_addr_base` attribute of the associated compilation unit. The address
2552 value A in the `.debug_addr` section has the size of the generic type.
2553
2554 It pushes a location description L with one memory location description SL
2555 on the stack. SL specifies the memory location storage corresponding to the
2556 target architecture default address space with a bit offset equal to A
2557 scaled by 8 (the byte size).
2558
2559 <i>If the DWARF is part of a code object, then A may need to be relocated.
2560 For example, in the ELF code object format, A must be adjusted by the
2561 difference between the ELF segment virtual address and the virtual address
2562 at which the segment is loaded.</i>
2563
2564 3. `DW_OP_form_tls_address`
2565
2566 `DW_OP_form_tls_address` pops one stack entry that must be an integral type
2567 value and treats it as a thread-local storage address TA.
2568
2569 It pushes a location description L with one memory location description SL
2570 on the stack. SL is the target architecture specific memory location
2571 description that corresponds to the thread-local storage address TA.
2572
2573 The meaning of the thread-local storage address TA is defined by the
2574 run-time environment. If the run-time environment supports multiple
2575 thread-local storage blocks for a single thread, then the block
2576 corresponding to the executable or shared library containing this DWARF
2577 expression is used.
2578
2579 <i>Some implementations of C, C++, Fortran, and other languages, support a
2580 thread-local storage class. Variables with this storage class have distinct
2581 values and addresses in distinct threads, much as automatic variables have
2582 distinct values and addresses in each subprogram invocation. Typically,
2583 there is a single block of storage containing all thread-local variables
2584 declared in the main executable, and a separate block for the variables
2585 declared in each shared library. Each thread-local variable can then be
2586 accessed in its block using an identifier. This identifier is typically a
2587 byte offset into the block and pushed onto the DWARF stack by one of the
2588 `DW_OP_const*` operations prior to the `DW_OP_form_tls_address` operation.
2589 Computing the address of the appropriate block can be complex (in some
2590 cases, the compiler emits a function call to do it), and difficult to
2591 describe using ordinary DWARF location descriptions. Instead of forcing
2592 complex thread-local storage calculations into the DWARF expressions, the
2593 `DW_OP_form_tls_address` allows the consumer to perform the computation
2594 based on the target architecture specific run-time environment.</i>
2595
2596 4. `DW_OP_call_frame_cfa`
2597
2598 `DW_OP_call_frame_cfa` pushes the location description L of the Canonical
2599 Frame Address (CFA) of the current subprogram, obtained from the call frame
2600 information on the stack. See [6.4 Call Frame
2601 Information](#call-frame-information).
2602
2603 <i>Although the value of the `DW_AT_frame_base` attribute of the debugger
2604 information entry corresponding to the current subprogram can be computed
2605 using a location list expression, in some cases this would require an
2606 extensive location list because the values of the registers used in
2607 computing the CFA change during a subprogram execution. If the call frame
2608 information is present, then it already encodes such changes, and it is
2609 space efficient to reference that using the `DW_OP_call_frame_cfa`
2610 operation.</i>
2611
2612 5. `DW_OP_fbreg`
2613
2614 `DW_OP_fbreg` has a single signed LEB128 integer operand that represents a
2615 byte displacement B.
2616
2617 The location description L for the <i>frame base</i> of the current
2618 subprogram is obtained from the `DW_AT_frame_base` attribute of the debugger
2619 information entry corresponding to the current subprogram as described in
2620 [3.3.5 Low-Level Information](#low-level-information).
2621
2622 The location description L is updated by bit offset B scaled by 8 (the byte
2623 size) and pushed on the stack.
2624
2625 6. `DW_OP_breg0`, `DW_OP_breg1`, ..., `DW_OP_breg31`
2626
2627 The `DW_OP_breg<N>` operations encode the numbers of up to 32 registers,
2628 numbered from 0 through 31, inclusive. The register number R corresponds to
2629 the N in the operation name.
2630
2631 They have a single signed LEB128 integer operand that represents a byte
2632 displacement B.
2633
2634 The address space identifier AS is defined as the one corresponding to the
2635 target architecture specific default address space.
2636
2637 The address size S is defined as the address bit size of the target
2638 architecture specific address space corresponding to AS.
2639
2640 The contents of the register specified by R are retrieved as if a
2641 `DW_OP_regval_type R, DR` operation was performed where DR is the offset of
2642 a hypothetical debug information entry in the current compilation unit for
2643 an unsigned integral base type of size S bits. B is added and the least
2644 significant S bits are treated as an unsigned value to be used as an address
2645 A.
2646
2647 They push a location description L comprising one memory location
2648 description LS on the stack. LS specifies the memory location storage that
2649 corresponds to AS with a bit offset equal to A scaled by 8 (the byte size).
2650
2651 7. `DW_OP_bregx`
2652
2653 `DW_OP_bregx` has two operands. The first is an unsigned LEB128 integer that
2654 represents a register number R. The second is a signed LEB128 integer that
2655 represents a byte displacement B.
2656
2657 The action is the same as for `DW_OP_breg<N>`, except that R is used as the
2658 register number and B is used as the byte displacement.
2659
2660 ###### A.2.5.4.4.4 Register Location Description Operations
2661
2662 > NOTE: This section replaces DWARF Version 5 section 2.6.1.1.3.
2663
2664 There is a register location storage that corresponds to each of the target
2665 architecture registers. The size of each register location storage corresponds
2666 to the size of the corresponding target architecture register.
2667
2668 A register location description specifies a register location storage. The bit
2669 offset corresponds to a bit position within the register. Bits accessed using a
2670 register location description access the corresponding target architecture
2671 register starting at the specified bit offset.
2672
2673 1. `DW_OP_reg0`, `DW_OP_reg1`, ..., `DW_OP_reg31`
2674
2675 `DW_OP_reg<N>` operations encode the numbers of up to 32 registers, numbered
2676 from 0 through 31, inclusive. The target architecture register number R
2677 corresponds to the N in the operation name.
2678
2679 The operation is equivalent to performing `DW_OP_regx R`.
2680
2681 2. `DW_OP_regx`
2682
2683 `DW_OP_regx` has a single unsigned LEB128 integer operand that represents a
2684 target architecture register number R.
2685
2686 If the current call frame is the top call frame, it pushes a location
2687 description L that specifies one register location description SL on the
2688 stack. SL specifies the register location storage that corresponds to R with
2689 a bit offset of 0 for the current thread.
2690
2691 If the current call frame is not the top call frame, call frame information
2692 (see [6.4 Call Frame Information](#call-frame-information)) is used to
2693 determine the location description that holds the register for the current
2694 call frame and current program location of the current thread. The resulting
2695 location description L is pushed.
2696
2697 <i>Note that if call frame information is used, the resulting location
2698 description may be register, memory, or undefined.</i>
2699
2700 <i>An implementation may evaluate the call frame information immediately, or
2701 may defer evaluation until L is accessed by an operation. If evaluation is
2702 deferred, R and the current context can be recorded in L. When accessed, the
2703 recorded context is used to evaluate the call frame information, not the
2704 current context of the access operation.</i>
2705
2706 <i>These operations obtain a register location. To fetch the contents of a
2707 register, it is necessary to use `DW_OP_regval_type`, use one of the
2708 `DW_OP_breg*` register-based addressing operations, or use `DW_OP_deref*` on a
2709 register location description.</i>
2710
2711 ###### A.2.5.4.4.5 Implicit Location Description Operations
2712
2713 > NOTE: This section replaces DWARF Version 5 section 2.6.1.1.4.
2714
2715 Implicit location storage represents a piece or all of an object which has no
2716 actual location in the program but whose contents are nonetheless known, either
2717 as a constant or can be computed from other locations and values in the program.
2718
2719 An implicit location description specifies an implicit location storage. The bit
2720 offset corresponds to a bit position within the implicit location storage. Bits
2721 accessed using an implicit location description, access the corresponding
2722 implicit storage value starting at the bit offset.
2723
2724 1. `DW_OP_implicit_value`
2725
2726 `DW_OP_implicit_value` has two operands. The first is an unsigned LEB128
2727 integer that represents a byte size S. The second is a block of bytes with a
2728 length equal to S treated as a literal value V.
2729
2730 An implicit location storage LS is created with the literal value V and a
2731 size of S.
2732
2733 It pushes location description L with one implicit location description SL
2734 on the stack. SL specifies LS with a bit offset of 0.
2735
2736 2. `DW_OP_stack_value`
2737
2738 `DW_OP_stack_value` pops one stack entry that must be a value V.
2739
2740 An implicit location storage LS is created with the literal value V using
2741 the size, encoding, and endianity specified by V's base type.
2742
2743 It pushes a location description L with one implicit location description SL
2744 on the stack. SL specifies LS with a bit offset of 0.
2745
2746 <i>The `DW_OP_stack_value` operation specifies that the object does not
2747 exist in memory, but its value is nonetheless known. In this form, the
2748 location description specifies the actual value of the object, rather than
2749 specifying the memory or register storage that holds the value.</i>
2750
2751 See `DW_OP_implicit_pointer` (following) for special rules concerning
2752 implicit pointer values produced by dereferencing implicit location
2753 descriptions created by the `DW_OP_implicit_pointer` operation.
2754
2755 Note: Since location descriptions are allowed on the stack, the
2756 `DW_OP_stack_value` operation no longer terminates the DWARF operation
2757 expression execution as in DWARF Version 5.
2758
2759 3. `DW_OP_implicit_pointer`
2760
2761 <i>An optimizing compiler may eliminate a pointer, while still retaining the
2762 value that the pointer addressed. `DW_OP_implicit_pointer` allows a producer
2763 to describe this value.</i>
2764
2765 <i>`DW_OP_implicit_pointer` specifies an object is a pointer to the target
2766 architecture default address space that cannot be represented as a real
2767 pointer, even though the value it would point to can be described. In this
2768 form, the location description specifies a debugging information entry that
2769 represents the actual location description of the object to which the
2770 pointer would point. Thus, a consumer of the debug information would be able
2771 to access the dereferenced pointer, even when it cannot access the pointer
2772 itself.</i>
2773
2774 `DW_OP_implicit_pointer` has two operands. The first operand is a 4-byte
2775 unsigned value in the 32-bit DWARF format, or an 8-byte unsigned value in
2776 the 64-bit DWARF format, that represents the byte offset DR of a debugging
2777 information entry D relative to the beginning of the `.debug_info` section
2778 that contains the current compilation unit. The second operand is a signed
2779 LEB128 integer that represents a byte displacement B.
2780
2781 <i>Note that D might not be in the current compilation unit.</i>
2782
2783 <i>The first operand interpretation is exactly like that for
2784 `DW_FORM_ref_addr`.</i>
2785
2786 The address space identifier AS is defined as the one corresponding to the
2787 target architecture specific default address space.
2788
2789 The address size S is defined as the address bit size of the target
2790 architecture specific address space corresponding to AS.
2791
2792 An implicit location storage LS is created with the debugging information
2793 entry D, address space AS, and size of S.
2794
2795 It pushes a location description L that comprises one implicit location
2796 description SL on the stack. SL specifies LS with a bit offset of 0.
2797
2798 It is an evaluation error if a `DW_OP_deref*` operation pops a location
2799 description L', and retrieves S bits, such that any retrieved bits come from
2800 an implicit location storage that is the same as LS, unless both the
2801 following conditions are met:
2802
2803 1. All retrieved bits come from an implicit location description that
2804 refers to an implicit location storage that is the same as LS.
2805
2806 <i>Note that all bits do not have to come from the same implicit
2807 location description, as L' may involve composite location
2808 descriptions.</i>
2809
2810 2. The bits come from consecutive ascending offsets within their respective
2811 implicit location storage.
2812
2813 <i>These rules are equivalent to retrieving the complete contents of LS.</i>
2814
2815 If both the above conditions are met, then the value V pushed by the
2816 `DW_OP_deref*` operation is an implicit pointer value IPV with a target
2817 architecture specific address space of AS, a debugging information entry of
2818 D, and a base type of T. If AS is the target architecture default address
2819 space, then T is the generic type. Otherwise, T is a target architecture
2820 specific integral type with a bit size equal to S.
2821
2822 If IPV is either implicitly converted to a location description (only done
2823 if AS is the target architecture default address space), then the resulting
2824 location description RL is:
2825
2826 - If D has a `DW_AT_location` attribute, the DWARF expression E from the
2827 `DW_AT_location` attribute is evaluated with the current context, except
2828 that the result kind is a location description, the compilation unit is
2829 the one that contains D, the object is unspecified, and the initial stack
2830 is empty. RL is the expression result.
2831
2832 <i>Note that E is evaluated with the context of the expression accessing
2833 IPV, and not the context of the expression that contained the
2834 `DW_OP_implicit_pointer` operation that created L.</i>
2835
2836 - If D has a `DW_AT_const_value` attribute, then an implicit location
2837 storage RLS is created from the `DW_AT_const_value` attribute's value with
2838 a size matching the size of the `DW_AT_const_value` attribute's value. RL
2839 comprises one implicit location description SRL. SRL specifies RLS with a
2840 bit offset of 0.
2841
2842 > NOTE: If using `DW_AT_const_value` for variables and formal parameters
2843 > is deprecated and instead `DW_AT_location` is used with an implicit
2844 > location description, then this rule would not be required.
2845
2846 - Otherwise, it is an evaluation error.
2847
2848 The location description RL is updated by bit offset B scaled by 8 (the byte
2849 size).
2850
2851 If a `DW_OP_stack_value` operation pops a value that is the same as IPV,
2852 then it pushes a location description that is the same as L.
2853
2854 It is an evaluation error if LS or IPV is accessed in any other manner.
2855
2856 <i>The restrictions on how an implicit pointer location description created
2857 by `DW_OP_implicit_pointer` can be used are to simplify the DWARF consumer.
2858 Similarly, for an implicit pointer value created by `DW_OP_deref*` and
2859 `DW_OP_stack_value`.</i>
2860
2861 <i>Typically a `DW_OP_implicit_pointer` operation is used in a DWARF expression
2862 E<sub>1</sub> of a `DW_TAG_variable` or `DW_TAG_formal_parameter` debugging
2863 information entry D<sub>1</sub>'s `DW_AT_location` attribute. The debugging
2864 information entry referenced by the `DW_OP_implicit_pointer` operation is
2865 typically itself a `DW_TAG_variable` or `DW_TAG_formal_parameter` debugging
2866 information entry D<sub>2</sub> whose `DW_AT_location` attribute gives a second
2867 DWARF expression E<sub>2</sub>.</i>
2868
2869 <i>D<sub>1</sub> and E<sub>1</sub> are describing the location of a pointer type
2870 object. D<sub>2</sub> and E<sub>2</sub> are describing the location of the
2871 object pointed to by that pointer object.</i>
2872
2873 <i>However, D<sub>2</sub> may be any debugging information entry that contains a
2874 `DW_AT_location` or `DW_AT_const_value` attribute (for example,
2875 `DW_TAG_dwarf_procedure`). By using E<sub>2</sub>, a consumer can reconstruct
2876 the value of the object when asked to dereference the pointer described by
2877 E<sub>1</sub> which contains the `DW_OP_implicit_pointer` operation.</i>
2878
2879 ###### A.2.5.4.4.6 Composite Location Description Operations
2880
2881 > NOTE: This section replaces DWARF Version 5 section 2.6.1.2.
2882
2883 A composite location storage represents an object or value which may be
2884 contained in part of another location storage or contained in parts of more than
2885 one location storage.
2886
2887 Each part has a part location description L and a part bit size S. L can have
2888 one or more single location descriptions SL. If there are more than one SL then
2889 that indicates that part is located in more than one place. The bits of each
2890 place of the part comprise S contiguous bits from the location storage LS
2891 specified by SL starting at the bit offset specified by SL. All the bits must be
2892 within the size of LS or the DWARF expression is ill-formed.
2893
2894 A composite location storage can have zero or more parts. The parts are
2895 contiguous such that the zero-based location storage bit index will range over
2896 each part with no gaps between them. Therefore, the size of a composite location
2897 storage is the sum of the size of its parts. The DWARF expression is ill-formed
2898 if the size of the contiguous location storage is larger than the size of the
2899 memory location storage corresponding to the largest target architecture
2900 specific address space.
2901
2902 A composite location description specifies a composite location storage. The bit
2903 offset corresponds to a bit position within the composite location storage.
2904
2905 There are operations that create a composite location storage.
2906
2907 There are other operations that allow a composite location storage to be
2908 incrementally created. Each part is created by a separate operation. There may
2909 be one or more operations to create the final composite location storage. A
2910 series of such operations describes the parts of the composite location storage
2911 that are in the order that the associated part operations are executed.
2912
2913 To support incremental creation, a composite location storage can be in an
2914 incomplete state. When an incremental operation operates on an incomplete
2915 composite location storage, it adds a new part.
2916
2917 A composite location description that specifies a composite location storage
2918 that is incomplete is termed an incomplete composite location description. A
2919 composite location description that specifies a composite location storage that
2920 is complete is termed a complete composite location description.
2921
2922 If the top stack entry is a location description that has one incomplete
2923 composite location description SL after the execution of an operation expression
2924 has completed, SL is converted to a complete composite location description.
2925
2926 <i>Note that this conversion does not happen after the completion of an
2927 operation expression that is evaluated on the same stack by the `DW_OP_call*`
2928 operations. Such executions are not a separate evaluation of an operation
2929 expression, but rather the continued evaluation of the same operation expression
2930 that contains the `DW_OP_call*` operation.</i>
2931
2932 If a stack entry is required to be a location description L, but L has an
2933 incomplete composite location description, then the DWARF expression is
2934 ill-formed. The exception is for the operations involved in incrementally
2935 creating a composite location description as described below.
2936
2937 <i>Note that a DWARF operation expression may arbitrarily compose composite
2938 location descriptions from any other location description, including those that
2939 have multiple single location descriptions, and those that have composite
2940 location descriptions.</i>
2941
2942 <i>The incremental composite location description operations are defined to be
2943 compatible with the definitions in DWARF Version 5.</i>
2944
2945 1. `DW_OP_piece`
2946
2947 `DW_OP_piece` has a single unsigned LEB128 integer that represents a byte
2948 size S.
2949
2950 The action is based on the context:
2951
2952 - If the stack is empty, then a location description L comprised of one
2953 incomplete composite location description SL is pushed on the stack.
2954
2955 An incomplete composite location storage LS is created with a single part
2956 P. P specifies a location description PL and has a bit size of S scaled by
2957 8 (the byte size). PL is comprised of one undefined location description
2958 PSL.
2959
2960 SL specifies LS with a bit offset of 0.
2961
2962 - Otherwise, if the top stack entry is a location description L comprised of
2963 one incomplete composite location description SL, then the incomplete
2964 composite location storage LS that SL specifies is updated to append a new
2965 part P. P specifies a location description PL and has a bit size of S
2966 scaled by 8 (the byte size). PL is comprised of one undefined location
2967 description PSL. L is left on the stack.
2968 - Otherwise, if the top stack entry is a location description or can be
2969 converted to one, then it is popped and treated as a part location
2970 description PL. Then:
2971
2972 - If the top stack entry (after popping PL) is a location description L
2973 comprised of one incomplete composite location description SL, then the
2974 incomplete composite location storage LS that SL specifies is updated to
2975 append a new part P. P specifies the location description PL and has a
2976 bit size of S scaled by 8 (the byte size). L is left on the stack.
2977 - Otherwise, a location description L comprised of one
2978 incomplete composite location description SL is pushed on
2979 the stack.
2980
2981 An incomplete composite location storage LS is created with a single
2982 part P. P specifies the location description PL and has a bit size of S
2983 scaled by 8 (the byte size).
2984
2985 SL specifies LS with a bit offset of 0.
2986
2987 - Otherwise, the DWARF expression is ill-formed
2988
2989 <i>Many compilers store a single variable in sets of registers or store a
2990 variable partially in memory and partially in registers. `DW_OP_piece`
2991 provides a way of describing where a part of a variable is located.</i>
2992
2993 <i>The evaluation rules for the `DW_OP_piece` operation allow it to be
2994 compatible with the DWARF Version 5 definition.</i>
2995
2996 > NOTE: Since these extensions allow location descriptions to be entries on
2997 > the stack, a simpler operation to create composite location descriptions
2998 > could be defined. For example, just one operation that specifies how many
2999 > parts, and pops pairs of stack entries for the part size and location
3000 > description. Not only would this be a simpler operation and avoid the
3001 > complexities of incomplete composite location descriptions, but it may
3002 > also have a smaller encoding in practice. However, the desire for
3003 > compatibility with DWARF Version 5 is likely a stronger consideration.
3004
3005 2. `DW_OP_bit_piece`
3006
3007 `DW_OP_bit_piece` has two operands. The first is an unsigned LEB128 integer
3008 that represents the part bit size S. The second is an unsigned LEB128
3009 integer that represents a bit displacement B.
3010
3011 The action is the same as for `DW_OP_piece`, except that any part created
3012 has the bit size S, and the location description PL of any created part is
3013 updated by a bit offset B.
3014
3015 <i>`DW_OP_bit_piece` is used instead of `DW_OP_piece` when the piece to be
3016 assembled is not byte-sized or is not at the start of the part location
3017 description.</i>
3018
3019 #### A.2.5.5 DWARF Location List Expressions
3020
3021 > NOTE: This section replaces DWARF Version 5 section 2.6.2.
3022
3023 <i>To meet the needs of recent computer architectures and optimization
3024 techniques, debugging information must be able to describe the location of an
3025 object whose location changes over the object's lifetime, and may reside at
3026 multiple locations during parts of an object's lifetime. Location list
3027 expressions are used in place of operation expressions whenever the object whose
3028 location is being described has these requirements.</i>
3029
3030 A location list expression consists of a series of location list entries. Each
3031 location list entry is one of the following kinds:
3032
3033 1. <i>Bounded location description</i>
3034
3035 This kind of location list entry provides an operation expression that
3036 evaluates to the location description of an object that is valid over a
3037 lifetime bounded by a starting and ending address. The starting address is
3038 the lowest address of the address range over which the location is valid.
3039 The ending address is the address of the first location past the highest
3040 address of the address range.
3041
3042 The location list entry matches when the current program location is within
3043 the given range.
3044
3045 There are several kinds of bounded location description entries which differ
3046 in the way that they specify the starting and ending addresses.
3047
3048 2. <i>Default location description</i>
3049
3050 This kind of location list entry provides an operation expression that
3051 evaluates to the location description of an object that is valid when no
3052 bounded location description entry applies.
3053
3054 The location list entry matches when the current program location is not
3055 within the range of any bounded location description entry.
3056
3057 3. <i>Base address</i>
3058
3059 This kind of location list entry provides an address to be used as the base
3060 address for beginning and ending address offsets given in certain kinds of
3061 bounded location description entries. The applicable base address of a
3062 bounded location description entry is the address specified by the closest
3063 preceding base address entry in the same location list. If there is no
3064 preceding base address entry, then the applicable base address defaults to
3065 the base address of the compilation unit (see DWARF Version 5 section
3066 3.1.1).
3067
3068 In the case of a compilation unit where all of the machine code is contained
3069 in a single contiguous section, no base address entry is needed.
3070
3071 4. <i>End-of-list</i>
3072
3073 This kind of location list entry marks the end of the location list
3074 expression.
3075
3076 The address ranges defined by the bounded location description entries of a
3077 location list expression may overlap. When they do, they describe a situation in
3078 which an object exists simultaneously in more than one place.
3079
3080 If all of the address ranges in a given location list expression do not
3081 collectively cover the entire range over which the object in question is
3082 defined, and there is no following default location description entry, it is
3083 assumed that the object is not available for the portion of the range that is
3084 not covered.
3085
3086 The result of the evaluation of a DWARF location list expression is:
3087
3088 - If the current program location is not specified, then it is an evaluation
3089 error.
3090
3091 > NOTE: If the location list only has a single default entry, should that be
3092 > considered a match if there is no program location? If there are non-default
3093 > entries then it seems it has to be an evaluation error when there is no
3094 > program location as that indicates the location depends on the program
3095 > location which is not known.
3096
3097 - If there are no matching location list entries, then the result is a location
3098 description that comprises one undefined location description.
3099 - Otherwise, the operation expression E of each matching location list entry is
3100 evaluated with the current context, except that the result kind is a location
3101 description, the object is unspecified, and the initial stack is empty. The
3102 location list entry result is the location description returned by the
3103 evaluation of E.
3104
3105 The result is a location description that is comprised of the union of the
3106 single location descriptions of the location description result of each
3107 matching location list entry.
3108
3109 A location list expression can only be used as the value of a debugger
3110 information entry attribute that is encoded using class `loclist` or
3111 `loclistsptr` (see [7.5.5 Classes and Forms](#classes-and-forms)). The value of
3112 the attribute provides an index into a separate object file section called
3113 `.debug_loclists` or `.debug_loclists.dwo` (for split DWARF object files) that
3114 contains the location list entries.
3115
3116 A `DW_OP_call*` and `DW_OP_implicit_pointer` operation can be used to specify a
3117 debugger information entry attribute that has a location list expression.
3118 Several debugger information entry attributes allow DWARF expressions that are
3119 evaluated with an initial stack that includes a location description that may
3120 originate from the evaluation of a location list expression.
3121
3122 <i>This location list representation, the `loclist` and `loclistsptr` class, and
3123 the related `DW_AT_loclists_base` attribute are new in DWARF Version 5. Together
3124 they eliminate most, or all of the code object relocations previously needed for
3125 location list expressions.</i>
3126
3127 > NOTE: The rest of this section is the same as DWARF Version 5 section 2.6.2.
3128
3129 ## A.3 Program Scope Entries
3130
3131 > NOTE: This section provides changes to existing debugger information entry
3132 > attributes. These would be incorporated into the corresponding DWARF Version 5
3133 > chapter 3 sections.
3134
3135 ### A.3.3 Subroutine and Entry Point Entries
3136
3137 #### A.3.3.5 Low-Level Information
3138
3139 1. A `DW_TAG_subprogram`, `DW_TAG_inlined_subroutine`, or `DW_TAG_entry_point`
3140 debugger information entry may have a `DW_AT_return_addr` attribute, whose
3141 value is a DWARF expression E.
3142
3143 The result of the attribute is obtained by evaluating E with a context that
3144 has a result kind of a location description, an unspecified object, the
3145 compilation unit that contains E, an empty initial stack, and other context
3146 elements corresponding to the source language thread of execution upon which
3147 the user is focused, if any. The result of the evaluation is the location
3148 description L of the place where the return address for the current call
3149 frame's subprogram or entry point is stored.
3150
3151 The DWARF is ill-formed if L is not comprised of one memory location
3152 description for one of the target architecture specific address spaces.
3153
3154 > NOTE: It is unclear why `DW_TAG_inlined_subroutine` has a
3155 > `DW_AT_return_addr` attribute but not a `DW_AT_frame_base` or
3156 > `DW_AT_static_link` attribute. Seems it would either have all of them or
3157 > none. Since inlined subprograms do not have a call frame it seems they
3158 > would have none of these attributes.
3159
3160 2. A `DW_TAG_subprogram` or `DW_TAG_entry_point` debugger information entry may
3161 have a `DW_AT_frame_base` attribute, whose value is a DWARF expression E.
3162
3163 The result of the attribute is obtained by evaluating E with a context that
3164 has a result kind of a location description, an unspecified object, the
3165 compilation unit that contains E, an empty initial stack, and other context
3166 elements corresponding to the source language thread of execution upon which
3167 the user is focused, if any.
3168
3169 The DWARF is ill-formed if E contains a `DW_OP_fbreg` operation, or the
3170 resulting location description L is not comprised of one single location
3171 description SL.
3172
3173 If SL is a register location description for register R, then L is replaced
3174 with the result of evaluating a `DW_OP_bregx R, 0` operation. This computes
3175 the frame base memory location description in the target architecture
3176 default address space.
3177
3178 <i>This allows the more compact `DW_OP_reg*` to be used instead of
3179 `DW_OP_breg* 0`.</i>
3180
3181 > NOTE: This rule could be removed and require the producer to create the
3182 > required location description directly using `DW_OP_call_frame_cfa` or
3183 > `DW_OP_breg*`. This would also then allow a target to implement the call
3184 > frames within a large register.
3185
3186 Otherwise, the DWARF is ill-formed if SL is not a memory location
3187 description in any of the target architecture specific address spaces.
3188
3189 The resulting L is the <i>frame base</i> for the subprogram or entry point.
3190
3191 <i>Typically, E will use the `DW_OP_call_frame_cfa` operation or be a stack
3192 pointer register plus or minus some offset.</i>
3193
3194 3. If a `DW_TAG_subprogram` or `DW_TAG_entry_point` debugger information entry
3195 is lexically nested, it may have a `DW_AT_static_link` attribute, whose
3196 value is a DWARF expression E.
3197
3198 The result of the attribute is obtained by evaluating E with a context that
3199 has a result kind of a location description, an unspecified object, the
3200 compilation unit that contains E, an empty initial stack, and other context
3201 elements corresponding to the source language thread of execution upon which
3202 the user is focused, if any. The result of the evaluation is the location
3203 description L of the <i>canonical frame address</i> (see [6.4 Call Frame
3204 Information](#call-frame-information)) of the relevant call frame of the
3205 subprogram instance that immediately lexically encloses the current call
3206 frame's subprogram or entry point.
3207
3208 The DWARF is ill-formed if L is is not comprised of one memory location
3209 description for one of the target architecture specific address spaces.
3210
3211 ### A.3.4 Call Site Entries and Parameters
3212
3213 #### A.3.4.2 Call Site Parameters
3214
3215 1. A `DW_TAG_call_site_parameter` debugger information entry may have a
3216 `DW_AT_call_value` attribute, whose value is a DWARF operation expression
3217 E<sub>1</sub>.
3218
3219 The result of the `DW_AT_call_value` attribute is obtained by evaluating
3220 E<sub>1</sub> with a context that has a result kind of a value, an unspecified
3221 object, the compilation unit that contains E, an empty initial stack, and other
3222 context elements corresponding to the source language thread of execution upon
3223 which the user is focused, if any. The resulting value V<sub>1</sub> is the
3224 value of the parameter at the time of the call made by the call site.
3225
3226 For parameters passed by reference, where the code passes a pointer to a
3227 location which contains the parameter, or for reference type parameters, the
3228 `DW_TAG_call_site_parameter` debugger information entry may also have a
3229 `DW_AT_call_data_location` attribute whose value is a DWARF operation expression
3230 E<sub>2</sub>, and a `DW_AT_call_data_value` attribute whose value is a DWARF
3231 operation expression E<sub>3</sub>.
3232
3233 The value of the `DW_AT_call_data_location` attribute is obtained by evaluating
3234 E<sub>2</sub> with a context that has a result kind of a location description,
3235 an unspecified object, the compilation unit that contains E, an empty initial
3236 stack, and other context elements corresponding to the source language thread of
3237 execution upon which the user is focused, if any. The resulting location
3238 description L<sub>2</sub> is the location where the referenced parameter lives
3239 during the call made by the call site. If E<sub>2</sub> would just be a
3240 `DW_OP_push_object_address`, then the `DW_AT_call_data_location` attribute may
3241 be omitted.
3242
3243 > NOTE: The DWARF Version 5 implies that `DW_OP_push_object_address` may be
3244 > used but does not state what object must be specified in the context.
3245 > Either `DW_OP_push_object_address` cannot be used, or the object to be
3246 > passed in the context must be defined.
3247
3248 The value of the `DW_AT_call_data_value` attribute is obtained by evaluating
3249 E<sub>3</sub> with a context that has a result kind of a value, an unspecified
3250 object, the compilation unit that contains E, an empty initial stack, and other
3251 context elements corresponding to the source language thread of execution upon
3252 which the user is focused, if any. The resulting value V<sub>3</sub> is the
3253 value in L<sub>2</sub> at the time of the call made by the call site.
3254
3255 The result of these attributes is undefined if the current call frame is not for
3256 the subprogram containing the `DW_TAG_call_site_parameter` debugger information
3257 entry or the current program location is not for the call site containing the
3258 `DW_TAG_call_site_parameter` debugger information entry in the current call
3259 frame.
3260
3261 <i>The consumer may have to virtually unwind to the call site (see [6.4 Call
3262 Frame Information](#call-frame-information)) in order to evaluate these
3263 attributes. This will ensure the source language thread of execution upon which
3264 the user is focused corresponds to the call site needed to evaluate the
3265 expression.</i>
3266
3267 If it is not possible to avoid the expressions of these attributes from
3268 accessing registers or memory locations that might be clobbered by the
3269 subprogram being called by the call site, then the associated attribute should
3270 not be provided.
3271
3272 <i>The reason for the restriction is that the parameter may need to be accessed
3273 during the execution of the callee. The consumer may virtually unwind from the
3274 called subprogram back to the caller and then evaluate the attribute
3275 expressions. The call frame information (see [6.4 Call Frame
3276 Information](#call-frame-information)) will not be able to restore registers
3277 that have been clobbered, and clobbered memory will no longer have the value at
3278 the time of the call.</i>
3279
3280 ### A.3.5 Lexical Block Entries
3281
3282 > NOTE: This section is the same as DWARF Version 5 section 3.5.
3283
3284 ## A.4 Data Object and Object List Entries
3285
3286 > NOTE: This section provides changes to existing debugger information entry
3287 > attributes. These would be incorporated into the corresponding DWARF Version 5
3288 > chapter 4 sections.
3289
3290 ### A.4.1 Data Object Entries
3291
3292 Program variables, formal parameters and constants are represented by debugging
3293 information entries with the tags `DW_TAG_variable`, `DW_TAG_formal_parameter`
3294 and `DW_TAG_constant`, respectively.
3295
3296 *The tag `DW_TAG_constant` is used for languages that have true named constants.*
3297
3298 The debugging information entry for a program variable, formal parameter or
3299 constant may have the following attributes:
3300
3301 1. A `DW_AT_location` attribute, whose value is a DWARF expression E that
3302 describes the location of a variable or parameter at run-time.
3303
3304 The result of the attribute is obtained by evaluating E with a context that
3305 has a result kind of a location description, an unspecified object, the
3306 compilation unit that contains E, an empty initial stack, and other context
3307 elements corresponding to the source language thread of execution upon which
3308 the user is focused, if any. The result of the evaluation is the location
3309 description of the base of the data object.
3310
3311 See [2.5.4.2 Control Flow Operations](#control-flow-operations) for special
3312 evaluation rules used by the `DW_OP_call*` operations.
3313
3314 > NOTE: Delete the description of how the `DW_OP_call*` operations evaluate
3315 > a `DW_AT_location` attribute as that is now described in the operations.
3316
3317 > NOTE: See the discussion about the `DW_AT_location` attribute in the
3318 > `DW_OP_call*` operation. Having each attribute only have a single purpose
3319 > and single execution semantics seems desirable. It makes it easier for the
3320 > consumer that no longer have to track the context. It makes it easier for
3321 > the producer as it can rely on a single semantics for each attribute.
3322 >
3323 > For that reason, limiting the `DW_AT_location` attribute to only
3324 > supporting evaluating the location description of an object, and using a
3325 > different attribute and encoding class for the evaluation of DWARF
3326 > expression <i>procedures</i> on the same operation expression stack seems
3327 > desirable.
3328
3329 2. `DW_AT_const_value`
3330
3331 > NOTE: Could deprecate using the `DW_AT_const_value` attribute for
3332 > `DW_TAG_variable` or `DW_TAG_formal_parameter` debugger information
3333 > entries that have been optimized to a constant. Instead, `DW_AT_location`
3334 > could be used with a DWARF expression that produces an implicit location
3335 > description now that any location description can be used within a DWARF
3336 > expression. This allows the `DW_OP_call*` operations to be used to push
3337 > the location description of any variable regardless of how it is
3338 > optimized.
3339
3340 ### A.4.2 Common Block Entries
3341
3342 A common block entry also has a DW_AT_location attribute whose value is a DWARF
3343 expression E that describes the location of the common block at run-time. The
3344 result of the attribute is obtained by evaluating E with a context that has a
3345 result kind of a location description, an unspecified object, the compilation
3346 unit that contains E, an empty initial stack, and other context elements
3347 corresponding to the source language thread of execution upon which the user is
3348 focused, if any. The result of the evaluation is the location description of the
3349 base of the common block. See 2.5.4.2 Control Flow Operations for special
3350 evaluation rules used by the DW_OP_call* operations.
3351
3352 ## A.5 Type Entries
3353
3354 > NOTE: This section provides changes to existing debugger information entry
3355 > attributes. These would be incorporated into the corresponding DWARF Version 5
3356 > chapter 5 sections.
3357
3358 ### A.5.7 Structure, Union, Class and Interface Type Entries
3359
3360 #### A.5.7.3 Derived or Extended Structures, Classes and Interfaces
3361
3362 1. For a `DW_AT_data_member_location` attribute there are two cases:
3363
3364 1. If the attribute is an integer constant B, it provides the offset in
3365 bytes from the beginning of the containing entity.
3366
3367 The result of the attribute is obtained by updating the bit offset of
3368 the location description of the beginning of the containing entity by B
3369 scaled by 8 (the byte size). The result is the location description of
3370 the base of the member entry.
3371
3372 <i>If the beginning of the containing entity is not byte aligned, then
3373 the beginning of the member entry has the same bit displacement within a
3374 byte.</i>
3375
3376 2. Otherwise, the attribute must be a DWARF expression E which is evaluated
3377 with a context that has a result kind of a location description, an
3378 unspecified object, the compilation unit that contains E, an initial
3379 stack comprising the location description of the beginning of the
3380 containing entity, and other context elements corresponding to the
3381 source language thread of execution upon which the user is focused, if
3382 any. The result of the evaluation is the location description of the
3383 base of the member entry.
3384
3385 > NOTE: The beginning of the containing entity can now be any location
3386 > description, including those with more than one single location
3387 > description, and those with single location descriptions that are of any
3388 > kind and have any bit offset.
3389
3390 #### A.5.7.8 Member Function Entries
3391
3392 1. An entry for a virtual function also has a `DW_AT_vtable_elem_location`
3393 attribute whose value is a DWARF expression E.
3394
3395 The result of the attribute is obtained by evaluating E with a context that
3396 has a result kind of a location description, an unspecified object, the
3397 compilation unit that contains E, an initial stack comprising the location
3398 description of the object of the enclosing type, and other context elements
3399 corresponding to the source language thread of execution upon which the user
3400 is focused, if any. The result of the evaluation is the location description
3401 of the slot for the function within the virtual function table for the
3402 enclosing class.
3403
3404 ### A.5.14 Pointer to Member Type Entries
3405
3406 1. The `DW_TAG_ptr_to_member_type` debugging information entry has a
3407 `DW_AT_use_location` attribute whose value is a DWARF expression E. It is used
3408 to compute the location description of the member of the class to which the
3409 pointer to member entry points.
3410
3411 <i>The method used to find the location description of a given member of a
3412 class, structure, or union is common to any instance of that class, structure,
3413 or union and to any instance of the pointer to member type. The method is thus
3414 associated with the pointer to member type, rather than with each object that
3415 has a pointer to member type.</i>
3416
3417 The `DW_AT_use_location` DWARF expression is used in conjunction with the
3418 location description for a particular object of the given pointer to member type
3419 and for a particular structure or class instance.
3420
3421 The result of the attribute is obtained by evaluating E with a context that has
3422 a result kind of a location description, an unspecified object, the compilation
3423 unit that contains E, an initial stack comprising two entries, and other context
3424 elements corresponding to the source language thread of execution upon which the
3425 user is focused, if any. The first stack entry is the value of the pointer to
3426 member object itself. The second stack entry is the location description of the
3427 base of the entire class, structure, or union instance containing the member
3428 whose location is being calculated. The result of the evaluation is the location
3429 description of the member of the class to which the pointer to member entry
3430 points.
3431
3432 ### A.5.18 Dynamic Properties of Types
3433
3434 #### A.5.18.1 Data Location
3435
3436 1. The `DW_AT_data_location` attribute may be used with any type that provides one
3437 or more levels of hidden indirection and/or run-time parameters in its
3438 representation. Its value is a DWARF operation expression E which computes the
3439 location description of the data for an object. When this attribute is omitted,
3440 the location description of the data is the same as the location description of
3441 the object.
3442
3443 The result of the attribute is obtained by evaluating E with a context that has
3444 a result kind of a location description, an object that is the location
3445 description of the data descriptor, the compilation unit that contains E, an
3446 empty initial stack, and other context elements corresponding to the source
3447 language thread of execution upon which the user is focused, if any. The result
3448 of the evaluation is the location description of the base of the member entry.
3449
3450 <i>E will typically involve an operation expression that begins with a
3451 `DW_OP_push_object_address` operation which loads the location description
3452 of the object which can then serve as a descriptor in subsequent
3453 calculation.</i>
3454
3455 > NOTE: Since `DW_AT_data_member_location`, `DW_AT_use_location`, and
3456 > `DW_AT_vtable_elem_location` allow both operation expressions and location
3457 > list expressions, why does `DW_AT_data_location` not allow both? In all cases
3458 > they apply to data objects so less likely that optimization would cause
3459 > different operation expressions for different program location ranges. But if
3460 > supporting for some then should be for all.
3461 >
3462 > It seems odd this attribute is not the same as `DW_AT_data_member_location` in
3463 > having an initial stack with the location description of the object since the
3464 > expression has to need it.
3465
3466 ## A.6 Other Debugging Information
3467
3468 > NOTE: This section provides changes to existing debugger information entry
3469 > attributes. These would be incorporated into the corresponding DWARF Version 5
3470 > chapter 6 sections.
3471
3472 ### A.6.2 Line Number Information
3473
3474 > NOTE: This section is the same as DWARF Version 5 section 6.2.
3475
3476 ### A.6.4 Call Frame Information
3477
3478 > NOTE: This section provides changes to DWARF Version 5 section 6.4. Register
3479 > unwind DWARF expressions are generalized to allow any location description,
3480 > including those with composite and implicit location descriptions.
3481
3482 #### A.6.4.1 Structure of Call Frame Information
3483
3484 The register rules are:
3485
3486 1. <i>undefined</i>
3487
3488 A register that has this rule has no recoverable value in the previous
3489 frame. The previous value of this register is the undefined location
3490 description (see [2.5.4.4.2 Undefined Location Description
3491 Operations](#undefined-location-description-operations)).
3492
3493 <i>By convention, the register is not preserved by a callee.</i>
3494
3495 2. <i>same value</i>
3496
3497 This register has not been modified from the previous caller frame.
3498
3499 If the current frame is the top frame, then the previous value of this
3500 register is the location description L that specifies one register location
3501 description SL. SL specifies the register location storage that corresponds
3502 to the register with a bit offset of 0 for the current thread.
3503
3504 If the current frame is not the top frame, then the previous value of this
3505 register is the location description obtained using the call frame
3506 information for the callee frame and callee program location invoked by the
3507 current caller frame for the same register.
3508
3509 <i>By convention, the register is preserved by the callee, but the callee
3510 has not modified it.</i>
3511
3512 3. <i>offset(N)</i>
3513
3514 N is a signed byte offset. The previous value of this register is saved at
3515 the location description L. Where L is the location description of the
3516 current CFA (see [2.5.4 DWARF Operation
3517 Expressions](#dwarf-operation-expressions)) updated with the bit offset N
3518 scaled by 8 (the byte size).
3519
3520 4. <i>val_offset(N)</i>
3521
3522 N is a signed byte offset. The previous value of this register is the memory
3523 byte address of the location description L. Where L is the location
3524 description of the current CFA (see [2.5.4 DWARF Operation
3525 Expressions](#dwarf-operation-expressions)) updated with the bit offset N
3526 scaled by 8 (the byte size).
3527
3528 The DWARF is ill-formed if the CFA location description is not a memory byte
3529 address location description, or if the register size does not match the
3530 size of an address in the target architecture default address space.
3531
3532 <i>Since the CFA location description is required to be a memory byte
3533 address location description, the value of val_offset(N) will also be a
3534 memory byte address location description since it is offsetting the CFA
3535 location description by N bytes. Furthermore, the value of val_offset(N)
3536 will be a memory byte address in the target architecture default address
3537 space.</i>
3538
3539 > NOTE: Should DWARF allow the address size to be a different size to the
3540 > size of the register? Requiring them to be the same bit size avoids any
3541 > issue of conversion as the bit contents of the register is simply
3542 > interpreted as a value of the address.
3543 >
3544 > GDB has a per register hook that allows a target specific conversion on a
3545 > register by register basis. It defaults to truncation of bigger registers,
3546 > and to actually reading bytes from the next register (or reads out of
3547 > bounds for the last register) for smaller registers. There are no GDB
3548 > tests that read a register out of bounds (except an illegal hand written
3549 > assembly test).
3550
3551 5. <i>register(R)</i>
3552
3553 This register has been stored in another register numbered R.
3554
3555 The previous value of this register is the location description obtained
3556 using the call frame information for the current frame and current program
3557 location for register R.
3558
3559 The DWARF is ill-formed if the size of this register does not match the size
3560 of register R or if there is a cyclic dependency in the call frame
3561 information.
3562
3563 > NOTE: Should this also allow R to be larger than this register? If so is
3564 > the value stored in the low order bits and it is undefined what is stored
3565 > in the extra upper bits?
3566
3567 6. <i>expression(E)</i>
3568
3569 The previous value of this register is located at the location description
3570 produced by evaluating the DWARF operation expression E (see [2.5.4 DWARF
3571 Operation Expressions](#dwarf-operation-expressions)).
3572
3573 E is evaluated with the current context, except the result kind is a
3574 location description, the compilation unit is unspecified, the object is
3575 unspecified, and an initial stack comprising the location description of the
3576 current CFA (see [2.5.4 DWARF Operation
3577 Expressions](#dwarf-operation-expressions)).
3578
3579 7. <i>val_expression(E)</i>
3580
3581 The previous value of this register is located at the implicit location
3582 description created from the value produced by evaluating the DWARF
3583 operation expression E (see [2.5.4 DWARF Operation
3584 Expressions](#dwarf-operation-expressions)).
3585
3586 E is evaluated with the current context, except the result kind is a value,
3587 the compilation unit is unspecified, the object is unspecified, and an
3588 initial stack comprising the location description of the current CFA (see
3589 [2.5.4 DWARF Operation Expressions](#dwarf-operation-expressions)).
3590
3591 The DWARF is ill-formed if the resulting value type size does not match the
3592 register size.
3593
3594 > NOTE: This has limited usefulness as the DWARF expression E can only
3595 > produce values up to the size of the generic type. This is due to not
3596 > allowing any operations that specify a type in a CFI operation expression.
3597 > This makes it unusable for registers that are larger than the generic
3598 > type. However, <i>expression(E)</i> can be used to create an implicit
3599 > location description of any size.
3600
3601 8. <i>architectural</i>
3602
3603 The rule is defined externally to this specification by the augmenter.
3604
3605 A Common Information Entry (CIE) holds information that is shared among many
3606 Frame Description Entries (FDE). There is at least one CIE in every non-empty
3607 `.debug_frame` section. A CIE contains the following fields, in order:
3608
3609 1. `length` (initial length)
3610
3611 A constant that gives the number of bytes of the CIE structure, not
3612 including the length field itself. The size of the length field plus the
3613 value of length must be an integral multiple of the address size specified
3614 in the `address_size` field.
3615
3616 2. `CIE_id` (4 or 8 bytes, see [7.4 32-Bit and 64-Bit DWARF Formats](#bit-and-64-bit-dwarf-formats))
3617
3618 A constant that is used to distinguish CIEs from FDEs.
3619
3620 In the 32-bit DWARF format, the value of the CIE id in the CIE header is
3621 0xffffffff; in the 64-bit DWARF format, the value is 0xffffffffffffffff.
3622
3623 3. `version` (ubyte)
3624
3625 A version number. This number is specific to the call frame information and
3626 is independent of the DWARF version number.
3627
3628 The value of the CIE version number is 4.
3629
3630 > NOTE: Would this be increased to 5 to reflect the changes in these
3631 > extensions?
3632
3633 4. `augmentation` (sequence of UTF-8 characters)
3634
3635 A null-terminated UTF-8 string that identifies the augmentation to this CIE
3636 or to the FDEs that use it. If a reader encounters an augmentation string
3637 that is unexpected, then only the following fields can be read:
3638
3639 - CIE: length, CIE_id, version, augmentation
3640 - FDE: length, CIE_pointer, initial_location, address_range
3641
3642 If there is no augmentation, this value is a zero byte.
3643
3644 <i>The augmentation string allows users to indicate that there is additional
3645 vendor and target architecture specific information in the CIE or FDE which
3646 is needed to virtually unwind a stack frame. For example, this might be
3647 information about dynamically allocated data which needs to be freed on exit
3648 from the routine.</i>
3649
3650 <i>Because the `.debug_frame` section is useful independently of any
3651 `.debug_info` section, the augmentation string always uses UTF-8
3652 encoding.</i>
3653
3654 5. `address_size` (ubyte)
3655
3656 The size of a target address in this CIE and any FDEs that use it, in bytes.
3657 If a compilation unit exists for this frame, its address size must match the
3658 address size here.
3659
3660 6. `segment_selector_size` (ubyte)
3661
3662 The size of a segment selector in this CIE and any FDEs that use it, in
3663 bytes.
3664
3665 7. `code_alignment_factor` (unsigned LEB128)
3666
3667 A constant that is factored out of all advance location instructions (see
3668 [6.4.2.1 Row Creation Instructions](#row-creation-instructions)). The
3669 resulting value is `(operand * code_alignment_factor)`.
3670
3671 8. `data_alignment_factor` (signed LEB128)
3672
3673 A constant that is factored out of certain offset instructions (see [6.4.2.2
3674 CFA Definition Instructions](#cfa-definition-instructions) and [6.4.2.3
3675 Register Rule Instructions](#register-rule-instructions)). The
3676 resulting value is `(operand * data_alignment_factor)`.
3677
3678 9. `return_address_register` (unsigned LEB128)
3679
3680 An unsigned LEB128 constant that indicates which column in the rule table
3681 represents the return address of the subprogram. Note that this column might
3682 not correspond to an actual machine register.
3683
3684 The value of the return address register is used to determine the program
3685 location of the caller frame. The program location of the top frame is the
3686 target architecture program counter value of the current thread.
3687
3688 10. `initial_instructions` (array of ubyte)
3689
3690 A sequence of rules that are interpreted to create the initial setting of
3691 each column in the table.
3692
3693 The default rule for all columns before interpretation of the initial
3694 instructions is the undefined rule. However, an ABI authoring body or a
3695 compilation system authoring body may specify an alternate default value for
3696 any or all columns.
3697
3698 11. `padding` (array of ubyte)
3699
3700 Enough `DW_CFA_nop` instructions to make the size of this entry match the
3701 length value above.
3702
3703 An FDE contains the following fields, in order:
3704
3705 1. `length` (initial length)
3706
3707 A constant that gives the number of bytes of the header and instruction
3708 stream for this subprogram, not including the length field itself. The size
3709 of the length field plus the value of length must be an integral multiple of
3710 the address size.
3711
3712 2. `CIE_pointer` (4 or 8 bytes, see [7.4 32-Bit and 64-Bit DWARF Formats](#bit-and-64-bit-dwarf-formats))
3713
3714 A constant offset into the `.debug_frame` section that denotes the CIE that
3715 is associated with this FDE.
3716
3717 3. `initial_location` (segment selector and target address)
3718
3719 The address of the first location associated with this table entry. If the
3720 segment_selector_size field of this FDE's CIE is non-zero, the initial
3721 location is preceded by a segment selector of the given length.
3722
3723 4. `address_range` (target address)
3724
3725 The number of bytes of program instructions described by this entry.
3726
3727 5. `instructions` (array of ubyte)
3728
3729 A sequence of table defining instructions that are described in [6.4.2 Call
3730 Frame Instructions](#call-frame-instructions).
3731
3732 6. `padding` (array of ubyte)
3733
3734 Enough `DW_CFA_nop` instructions to make the size of this entry match the
3735 length value above.
3736
3737 #### A.6.4.2 Call Frame Instructions
3738
3739 Some call frame instructions have operands that are encoded as DWARF operation
3740 expressions E (see [2.5.4 DWARF Operation
3741 Expressions](#dwarf-operation-expressions)). The DWARF operations that can be
3742 used in E have the following restrictions:
3743
3744 - `DW_OP_addrx`, `DW_OP_call2`, `DW_OP_call4`, `DW_OP_call_ref`,
3745 `DW_OP_const_type`, `DW_OP_constx`, `DW_OP_convert`, `DW_OP_deref_type`,
3746 `DW_OP_fbreg`, `DW_OP_implicit_pointer`, `DW_OP_regval_type`,
3747 `DW_OP_reinterpret`, and `DW_OP_xderef_type` operations are not allowed
3748 because the call frame information must not depend on other debug sections.
3749 - `DW_OP_push_object_address` is not allowed because there is no object context
3750 to provide a value to push.
3751 - `DW_OP_call_frame_cfa` and `DW_OP_entry_value` are not allowed because their
3752 use would be circular.
3753
3754 <i>Call frame instructions to which these restrictions apply include
3755 `DW_CFA_def_cfa_expression`, `DW_CFA_expression`, and
3756 `DW_CFA_val_expression`.</i>
3757
3758 ##### A.6.4.2.1 Row Creation Instructions
3759
3760 > NOTE: These instructions are the same as in DWARF Version 5 section 6.4.2.1.
3761
3762 ##### A.6.4.2.2 CFA Definition Instructions
3763
3764 1. `DW_CFA_def_cfa`
3765
3766 The `DW_CFA_def_cfa` instruction takes two unsigned LEB128 operands
3767 representing a register number R and a (non-factored) byte displacement B.
3768 The required action is to define the current CFA rule to be equivalent to
3769 the result of evaluating the DWARF operation expression `DW_OP_bregx R, B`
3770 as a location description.
3771
3772 2. `DW_CFA_def_cfa_sf`
3773
3774 The `DW_CFA_def_cfa_sf` instruction takes two operands: an unsigned LEB128
3775 value representing a register number R and a signed LEB128 factored byte
3776 displacement B. The required action is to define the current CFA rule to be
3777 equivalent to the result of evaluating the DWARF operation expression
3778 `DW_OP_bregx R, B * data_alignment_factor` as a location description.
3779
3780 <i>The action is the same as `DW_CFA_def_cfa`, except that the second
3781 operand is signed and factored.</i>
3782
3783 3. `DW_CFA_def_cfa_register`
3784
3785 The `DW_CFA_def_cfa_register` instruction takes a single unsigned LEB128
3786 operand representing a register number R. The required action is to define
3787 the current CFA rule to be equivalent to the result of evaluating the DWARF
3788 operation expression `DW_OP_bregx R, B` as a location description. B is the
3789 old CFA byte displacement.
3790
3791 If the subprogram has no current CFA rule, or the rule was defined by a
3792 `DW_CFA_def_cfa_expression` instruction, then the DWARF is ill-formed.
3793
3794 4. `DW_CFA_def_cfa_offset`
3795
3796 The `DW_CFA_def_cfa_offset` instruction takes a single unsigned LEB128
3797 operand representing a (non-factored) byte displacement B. The required
3798 action is to define the current CFA rule to be equivalent to the result of
3799 evaluating the DWARF operation expression `DW_OP_bregx R, B` as a location
3800 description. R is the old CFA register number.
3801
3802 If the subprogram has no current CFA rule, or the rule was defined by a
3803 `DW_CFA_def_cfa_expression` instruction, then the DWARF is ill-formed.
3804
3805 5. `DW_CFA_def_cfa_offset_sf`
3806
3807 The `DW_CFA_def_cfa_offset_sf` instruction takes a signed LEB128 operand
3808 representing a factored byte displacement B. The required action is to
3809 define the current CFA rule to be equivalent to the result of evaluating the
3810 DWARF operation expression `DW_OP_bregx R, B * data_alignment_factor` as a
3811 location description. R is the old CFA register number.
3812
3813 If the subprogram has no current CFA rule, or the rule was defined by a
3814 `DW_CFA_def_cfa_expression` instruction, then the DWARF is ill-formed.
3815
3816 <i>The action is the same as `DW_CFA_def_cfa_offset`, except that the
3817 operand is signed and factored.</i>
3818
3819 6. `DW_CFA_def_cfa_expression`
3820
3821 The `DW_CFA_def_cfa_expression` instruction takes a single operand encoded
3822 as a `DW_FORM_exprloc` value representing a DWARF operation expression E.
3823 The required action is to define the current CFA rule to be equivalent to
3824 the result of evaluating E with the current context, except the result kind
3825 is a location description, the compilation unit is unspecified, the object
3826 is unspecified, and an empty initial stack.
3827
3828 <i>See [6.4.2 Call Frame Instructions](#call-frame-instructions) regarding
3829 restrictions on the DWARF expression operations that can be used in E.</i>
3830
3831 The DWARF is ill-formed if the result of evaluating E is not a memory byte
3832 address location description.
3833
3834 ##### A.6.4.2.3 Register Rule Instructions
3835
3836 1. `DW_CFA_undefined`
3837
3838 The `DW_CFA_undefined` instruction takes a single unsigned LEB128 operand
3839 that represents a register number R. The required action is to set the rule
3840 for the register specified by R to `undefined`.
3841
3842 2. `DW_CFA_same_value`
3843
3844 The `DW_CFA_same_value` instruction takes a single unsigned LEB128 operand
3845 that represents a register number R. The required action is to set the rule
3846 for the register specified by R to `same value`.
3847
3848 3. `DW_CFA_offset`
3849
3850 The `DW_CFA_offset` instruction takes two operands: a register number R
3851 (encoded with the opcode) and an unsigned LEB128 constant representing a
3852 factored displacement B. The required action is to change the rule for the
3853 register specified by R to be an <i>offset(B * data_alignment_factor)</i>
3854 rule.
3855
3856 > NOTE: Seems this should be named `DW_CFA_offset_uf` since the offset is
3857 > unsigned factored.
3858
3859 4. `DW_CFA_offset_extended`
3860
3861 The `DW_CFA_offset_extended` instruction takes two unsigned LEB128 operands
3862 representing a register number R and a factored displacement B. This
3863 instruction is identical to `DW_CFA_offset`, except for the encoding and
3864 size of the register operand.
3865
3866 > NOTE: Seems this should be named `DW_CFA_offset_extended_uf` since the
3867 > displacement is unsigned factored.
3868
3869 5. `DW_CFA_offset_extended_sf`
3870
3871 The `DW_CFA_offset_extended_sf` instruction takes two operands: an unsigned
3872 LEB128 value representing a register number R and a signed LEB128 factored
3873 displacement B. This instruction is identical to `DW_CFA_offset_extended`,
3874 except that B is signed.
3875
3876 6. `DW_CFA_val_offset`
3877
3878 The `DW_CFA_val_offset` instruction takes two unsigned LEB128 operands
3879 representing a register number R and a factored displacement B. The required
3880 action is to change the rule for the register indicated by R to be a
3881 <i>val_offset(B * data_alignment_factor)</i> rule.
3882
3883 > NOTE: Seems this should be named `DW_CFA_val_offset_uf` since the
3884 displacement is unsigned factored.
3885
3886 7. `DW_CFA_val_offset_sf`
3887
3888 The `DW_CFA_val_offset_sf` instruction takes two operands: an unsigned
3889 LEB128 value representing a register number R and a signed LEB128 factored
3890 displacement B. This instruction is identical to `DW_CFA_val_offset`, except
3891 that B is signed.
3892
3893 8. `DW_CFA_register`
3894
3895 The `DW_CFA_register` instruction takes two unsigned LEB128 operands
3896 representing register numbers R1 and R2 respectively. The required action is
3897 to set the rule for the register specified by R1 to be a <i>register(R2)</i>
3898 rule.
3899
3900 9. `DW_CFA_expression`
3901
3902 The `DW_CFA_expression` instruction takes two operands: an unsigned LEB128
3903 value representing a register number R, and a `DW_FORM_block` value
3904 representing a DWARF operation expression E. The required action is to
3905 change the rule for the register specified by R to be an
3906 <i>expression(E)</i> rule.
3907
3908 <i>That is, E computes the location description where the register value can
3909 be retrieved.</i>
3910
3911 <i>See [6.4.2 Call Frame Instructions](#call-frame-instructions) regarding
3912 restrictions on the DWARF expression operations that can be used in E.</i>
3913
3914 10. `DW_CFA_val_expression`
3915
3916 The `DW_CFA_val_expression` instruction takes two operands: an unsigned
3917 LEB128 value representing a register number R, and a `DW_FORM_block` value
3918 representing a DWARF operation expression E. The required action is to
3919 change the rule for the register specified by R to be a
3920 <i>val_expression(E)</i> rule.
3921
3922 <i>That is, E computes the value of register R.</i>
3923
3924 <i>See [6.4.2 Call Frame Instructions](#call-frame-instructions) regarding
3925 restrictions on the DWARF expression operations that can be used in E.</i>
3926
3927 If the result of evaluating E is not a value with a base type size that
3928 matches the register size, then the DWARF is ill-formed.
3929
3930 11. `DW_CFA_restore`
3931
3932 The `DW_CFA_restore` instruction takes a single operand (encoded with the
3933 opcode) that represents a register number R. The required action is to
3934 change the rule for the register specified by R to the rule assigned it by
3935 the `initial_instructions` in the CIE.
3936
3937 12. `DW_CFA_restore_extended`
3938
3939 The `DW_CFA_restore_extended` instruction takes a single unsigned LEB128
3940 operand that represents a register number R. This instruction is identical
3941 to `DW_CFA_restore`, except for the encoding and size of the register
3942 operand.
3943
3944 ##### A.6.4.2.4 Row State Instructions
3945
3946 > NOTE: These instructions are the same as in DWARF Version 5 section 6.4.2.4.
3947
3948 ##### A.6.4.2.5 Padding Instruction
3949
3950 > NOTE: These instructions are the same as in DWARF Version 5 section 6.4.2.5.
3951
3952 #### A.6.4.3 Call Frame Instruction Usage
3953
3954 > NOTE: The same as in DWARF Version 5 section 6.4.3.
3955
3956 #### A.6.4.4 Call Frame Calling Address
3957
3958 > NOTE: The same as in DWARF Version 5 section 6.4.4.
3959
3960 ## A.7 Data Representation
3961
3962 > NOTE: This section provides changes to existing debugger information entry
3963 > attributes. These would be incorporated into the corresponding DWARF Version 5
3964 > chapter 7 sections.
3965
3966 ### A.7.4 32-Bit and 64-Bit DWARF Formats
3967
3968 > NOTE: This augments DWARF Version 5 section 7.4 list item 3's table.
3969
3970 Form Role
3971 ------------------------ --------------------------------------
3972 DW_OP_implicit_pointer offset in `.debug_info`
3973
3974 ### A.7.5 Format of Debugging Information
3975
3976 #### A.7.5.5 Classes and Forms
3977
3978 > NOTE: The same as in DWARF Version 5 section 7.5.5.
3979
3980 ### A.7.7 DWARF Expressions
3981
3982 > NOTE: Rename DWARF Version 5 section 7.7 to reflect the unification of
3983 > location descriptions into DWARF expressions.
3984
3985 #### A.7.7.1 Operation Expressions
3986
3987 > NOTE: Rename DWARF Version 5 section 7.7.1 and delete section 7.7.2 to reflect
3988 > the unification of location descriptions into DWARF expressions.
3989
3990 #### A.7.7.3 Location List Expressions
3991
3992 > NOTE: Rename DWARF Version 5 section 7.7.3 to reflect that location lists are
3993 > a kind of DWARF expression.
3994
3995 # B. Further Information
3996
3997 The following references provide additional information on the extension.
3998
3999 A reference to the DWARF standard is provided.
4000
4001 A formatted version of this extension is available on the LLVM site. It includes
4002 many figures that help illustrate the textual description, especially of the
4003 example DWARF expression evaluations.
4004
4005 Slides and a video of a presentation at the Linux Plumbers Conference 2021
4006 related to this extension are available.
4007
4008 The LLVM compiler extension includes the operations mentioned in the motivating
4009 examples. It also covers other extensions needed for heterogeneous devices.
4010
4011 - [DWARF Debugging Information Format](https://dwarfstd.org/)
4012 - [DWARF Debugging Information Format Version 5](https://dwarfstd.org/Dwarf5Std.php)
4013 - [Allow Location Descriptions on the DWARF Expression Stack](https://llvm.org/docs/AMDGPUDwarfExtensionAllowLocationDescriptionOnTheDwarfExpressionStack/AMDGPUDwarfExtensionAllowLocationDescriptionOnTheDwarfExpressionStack.html)
4014 - DWARF extensions for optimized SIMT/SIMD (GPU) debugging - Linux Plumbers Conference 2021
4015 - [Video](https://www.youtube.com/watch?v=QiR0ra0ymEY&t=10015s)
4016 - [Slides](https://linuxplumbersconf.org/event/11/contributions/1012/attachments/798/1505/DWARF_Extensions_for_Optimized_SIMT-SIMD_GPU_Debugging-LPC2021.pdf)
4017 - [DWARF Extensions For Heterogeneous Debugging](https://llvm.org/docs/AMDGPUDwarfExtensionsForHeterogeneousDebugging.html)