[dwarfdump] Add verifier check to find DIEs with DW_CHILDREN_yes but without children.

[JDevlieghere]

Summary

Flag DIEs that have DW_CHILDREN_yes set in their abbreviation but don't actually have any children.

rdar://59809554

[dblaikie]

does this classify as a warning? Is there a distinction between ill-formed DWARF and "this is valid, but unlikely to be great"?

[JDevlieghere]

In D88048#2286251, @dblaikie wrote:

does this classify as a warning? Is there a distinction between ill-formed DWARF and "this is valid, but unlikely to be great"?

I was on the fence myself, I couldn't find anything conclusive in the standard. I've made it a warning.

[dblaikie]

maybe slightly long test case - could do this with the CU DIE itself? (perhaps a case where hand-crafted/substantially hand-modified assembly is OK, because it can be made /so/ simple - or maybe yaml2obj dwarf support is adequate for this use case?)

[Higuoxing]

In D88048#2286589, @dblaikie wrote:

maybe slightly long test case - could do this with the CU DIE itself? (perhaps a case where hand-crafted/substantially hand-modified assembly is OK, because it can be made /so/ simple - or maybe yaml2obj dwarf support is adequate for this use case?)

Hi, currently, obj2yaml doesn't work properly when dumping the .debug_abbrev/__debug_abbrev section. We might not be able to get the correct output when there are multiple DIEs in one abbrev table. D87179 has fixed it but hasn't been landed yet. I've created one test case for this patch. Hope it works :-)

--- !ELF
FileHeader:
  Class: ELFCLASS64
  Data:  ELFDATA2LSB
  Type:  ET_EXEC
DWARF:
  debug_abbrev:
    - Table:
        - Tag:      DW_TAG_compile_unit
          Children: DW_CHILDREN_yes
          Attributes:
            - Attribute: DW_AT_low_pc
              Form:      DW_FORM_data4
  debug_info:
    - Version: 4
      Entries:
        - AbbrCode: 1
          Values:
            - Value: 0x1234
        - AbbrCode: 0   ## Terminator for the current DIE.

However, it seems that the assertion isValid() && "must check validity prior to calling" is triggered when the DIE isn't terminated.

--- !ELF
FileHeader:
  Class: ELFCLASS64
  Data:  ELFDATA2LSB
  Type:  ET_EXEC
DWARF:
  debug_abbrev:
    - Table:
        - Tag:      DW_TAG_compile_unit
          Children: DW_CHILDREN_yes
          Attributes:
            - Attribute: DW_AT_low_pc
              Form:      DW_FORM_data4
  debug_info:
    - Version: 4
      Entries:
        - AbbrCode: 1
          Values:
            - Value: 0x1234

See my inline comments.

[Higuoxing]

llvm/lib/DebugInfo/DWARF/DWARFVerifier.cpp

176

[aprantl]

Warning seems fine to me since it's more an inefficient encoding (an empty list of children) than malformed.

LGTM once everyone else is happy.

[JDevlieghere]

In D88048#2286776, @Higuoxing wrote:

In D88048#2286589, @dblaikie wrote:

maybe slightly long test case - could do this with the CU DIE itself? (perhaps a case where hand-crafted/substantially hand-modified assembly is OK, because it can be made /so/ simple - or maybe yaml2obj dwarf support is adequate for this use case?)

Hi, currently, obj2yaml doesn't work properly when dumping the .debug_abbrev/__debug_abbrev section. We might not be able to get the correct output when there are multiple DIEs in one abbrev table. D87179 has fixed it but hasn't been landed yet. I've created one test case for this patch. Hope it works :-)

(I forgot to submit the comment before landing this)

Awesome, thanks a lot, that's a great test case! :-)

llvm/lib/DebugInfo/DWARF/DWARFVerifier.cpp

for (unsigned I = 0; I < NumDies; ++I) {
     if (Die.getTag() == DW_TAG_null)
       continue;
 
     for (auto AttrValue : Die.attributes()) {
       NumUnitErrors += verifyDebugInfoAttribute(Die, AttrValue);
       NumUnitErrors += verifyDebugInfoForm(Die, AttrValue);
     }
 
+    if (Die.hasChildren()) {
+      if (Die.getFirstChild().isValid() &&
+          Die.getFirstChild().getTag() == DW_TAG_null) {
+        warn() << dwarf::TagString(Die.getTag())
+               << " has DW_CHILDREN_yes but DIE has no children: ";
+        Die.dump(OS);
+      }
+    }
+
     NumUnitErrors += verifyDebugInfoCallSite(Die);
   }
 
   DWARFDie Die = Unit.getUnitDIE(/* ExtractUnitDIEOnly = */ false);
   if (!Die) {
     error() << "Compilation unit without DIE.\n";
     NumUnitErrors++;
     return NumUnitErrors;

llvm/test/DebugInfo/X86/skeleton-unit-verify.s

 # RUN: llvm-mc -triple x86_64-unknown-linux %s -filetype=obj -o %t.o
 # RUN: not llvm-dwarfdump --verify %t.o | FileCheck %s
 
 # CHECK: Verifying .debug_abbrev...
 # CHECK-NEXT: Verifying .debug_info Unit Header Chain...
+# CHECK-NEXT: warning: DW_TAG_skeleton_unit has DW_CHILDREN_yes but DIE has no children
+# CHECK-NEXT: DW_TAG_skeleton_unit
 # CHECK-NEXT: error: Skeleton compilation unit has children.
 # CHECK-NEXT: Verifying .debug_info references...
 # CHECK-NEXT: Verifying .debug_types Unit Header Chain...
 # CHECK-NEXT: Errors detected.
 
         .section .debug_abbrev,"",@progbits
         .byte   1                       # Abbreviation Code
         .byte   74                      # DW_TAG_skeleton_unit

 .Lcu_begin0:
         .long   .Lcu_end0-.Lcu_start0 # Length of Unit
 .Lcu_start0:
         .short  5                       # DWARF version number
         .byte   4                       # DWARF Unit Type
         .byte   8                       # Address Size (in bytes)
         .long   .debug_abbrev           # Offset Into Abbrev. Section
         .quad   -6573227469967412476
-        .byte   1                       # Abbrev [1] 
+        .byte   1                       # Abbrev [1]
         .byte   0
 .Lcu_end0:
         .long   .Lcu_end1-.Lcu_start1 # Length of Unit
 .Lcu_start1:
         .short  5                       # DWARF version number
         .byte   4                       # DWARF Unit Type
         .byte   8                       # Address Size (in bytes)
         .long   .debug_abbrev           # Offset Into Abbrev. Section
         .quad   -6573227469967412476
         .byte   2                       # Abbrev [2]
         .byte   0
 .Lcu_end1:
 
 

llvm/test/tools/llvm-dwarfdump/X86/no-children.yaml

+# RUN: yaml2obj %s | llvm-dwarfdump -verify - | FileCheck %s
+# CHECK: warning: DW_TAG_compile_unit has DW_CHILDREN_yes but DIE has no children
+
+--- !ELF
+FileHeader:
+  Class: ELFCLASS64
+  Data:  ELFDATA2LSB
+  Type:  ET_EXEC
+DWARF:
+  debug_abbrev:
+    - Table:
+        - Tag:      DW_TAG_compile_unit
+          Children: DW_CHILDREN_yes
+          Attributes:
+            - Attribute: DW_AT_low_pc
+              Form:      DW_FORM_data4
+  debug_info:
+    - Version: 4
+      Entries:
+        - AbbrCode: 1
+          Values:
+            - Value: 0x1234
+        - AbbrCode: 0   ## Terminator for the current DIE.

llvm/test/tools/llvm-dwarfdump/X86/verify_die_ranges.s

 # RUN: llvm-mc %s -filetype obj -triple x86_64-apple-darwin -o - \
 # RUN: | not llvm-dwarfdump -verify - \
 # RUN: | FileCheck %s
 
 # CHECK: Verifying .debug_info Unit Header Chain...
-# CHECK-NEXT: error: Invalid address range [0x0000000000000007, 0x0000000000000006)
+# CHECK: warning: DW_TAG_compile_unit has DW_CHILDREN_yes but DIE has no children:
+# CHECK: error: Invalid address range [0x0000000000000007, 0x0000000000000006)
 
   .section  __TEXT,__text,regular,pure_instructions
   .macosx_version_min 10, 12
   .globl  _foo                    ## -- Begin function foo
   .p2align  4, 0x90
 _foo:                                   ## @foo
 Lfunc_begin0:
   .file 1 "basic.c"