Skip to content

Commit

Permalink
[DWARFVerifier] Allow overlapping ranges for ICF-merged functions (ll…
Browse files Browse the repository at this point in the history
…vm#117952)

This patch modifies the DWARF verifier to handle a valid case where two
or more functions have identical address ranges due to being merged by
ICF (Identical Code Folding). Previously, the verifier would incorrectly
report these as errors, but functions merged via ICF (such as when using
LLD's --keep-icf-stabs option) can legitimately share the same address
range.

A new test case has been added to verify this behavior using YAML-based
DWARF data that simulates two DW_TAG_subprogram entries with identical
address ranges. The test ensures that the verifier correctly identifies
this as a valid case and doesn't emit any errors, while still
maintaining the existing verification for truly invalid overlapping
ranges in other scenarios. Before this change, the newly added test case
would have failed, with `llvm-dwarfdump` marking the overlapping address
ranges in the DWARF as an error.

We also modify the existing tests `llvm-dwarfutil/ELF/X86/verify.test` and 
`llvm/test/tools/llvm-dwarfdump/X86/verify_parent_zero_length.yaml`
which rely on the existence of the error that we're trying to
suppress. We slightly change one offset so that the ranges don't
perfectly overlap and an error is still generated.
  • Loading branch information
alx32 authored Dec 17, 2024
1 parent de2acda commit ad32576
Show file tree
Hide file tree
Showing 5 changed files with 175 additions and 10 deletions.
10 changes: 7 additions & 3 deletions llvm/include/llvm/DebugInfo/DWARF/DWARFVerifier.h
Original file line number Diff line number Diff line change
Expand Up @@ -68,7 +68,9 @@ class DWARFVerifier {

/// Inserts the address range. If the range overlaps with an existing
/// range, the range that it overlaps with will be returned and the two
/// address ranges will be unioned together in "Ranges".
/// address ranges will be unioned together in "Ranges". If a duplicate
/// entry is attempted to be added, the duplicate range will not actually be
/// added and the returned iterator will point to end().
///
/// This is used for finding overlapping ranges in the DW_AT_ranges
/// attribute of a DIE. It is also used as a set of address ranges that
Expand All @@ -77,7 +79,9 @@ class DWARFVerifier {

/// Inserts the address range info. If any of its ranges overlaps with a
/// range in an existing range info, the range info is *not* added and an
/// iterator to the overlapping range info.
/// iterator to the overlapping range info. If a duplicate entry is
/// attempted to be added, the duplicate range will not actually be added
/// and the returned iterator will point to end().
///
/// This is used for finding overlapping children of the same DIE.
die_range_info_iterator insert(const DieRangeInfo &RI);
Expand All @@ -86,7 +90,7 @@ class DWARFVerifier {
bool contains(const DieRangeInfo &RHS) const;

/// Return true if any range in this object intersects with any range in
/// RHS.
/// RHS. Identical ranges are not considered to be intersecting.
bool intersects(const DieRangeInfo &RHS) const;
};

Expand Down
12 changes: 10 additions & 2 deletions llvm/lib/DebugInfo/DWARF/DWARFVerifier.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,11 @@ DWARFVerifier::DieRangeInfo::insert(const DWARFAddressRange &R) {
auto End = Ranges.end();
auto Pos = std::lower_bound(Begin, End, R);

// Check for exact duplicates which is an allowed special case
if (Pos != End && *Pos == R) {
return std::nullopt;
}

if (Pos != End) {
DWARFAddressRange Range(*Pos);
if (Pos->merge(R))
Expand Down Expand Up @@ -113,8 +118,11 @@ bool DWARFVerifier::DieRangeInfo::intersects(const DieRangeInfo &RHS) const {
auto I1 = Ranges.begin(), E1 = Ranges.end();
auto I2 = RHS.Ranges.begin(), E2 = RHS.Ranges.end();
while (I1 != E1 && I2 != E2) {
if (I1->intersects(*I2))
return true;
if (I1->intersects(*I2)) {
// Exact duplicates are allowed
if (!(*I1 == *I2))
return true;
}
if (I1->LowPC < I2->LowPC)
++I1;
else
Expand Down
153 changes: 153 additions & 0 deletions llvm/test/tools/llvm-dwarfdump/X86/verify_no_overlap_error_icf.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,153 @@
#--- comments.txt

# This test verifies several scenarios with DW_TAG_subprogram address ranges:
# 1. Two subprograms can have identical ranges (shown with foo2 and foo3 having same low_pc/high_pc)
# This is valid and can happen when ICF (Identical Code Folding) merges functions.
# 2. Two subprograms can have overlapping ranges when using DW_AT_ranges
# (shown with func1_with_ranges and func2_with_ranges sharing range 0x5000-0x6000)
# This is also valid and can occur with -fbasic-block-sections=all
# 3. The test also verifies that non-identical overlapping ranges are correctly flagged as errors:
# - When modifying just the first range's high offset from 0x6000 to 0x5999, it creates an invalid subrange overlap
# - When modifying just the first instance of DW_AT_high_pc 0x77 to 0x66, it creates an invalid function overlap
# The test ensures llvm-dwarfdump --verify correctly validates these cases by:
# a) Accepting valid identical overlapping ranges
# b) Rejecting invalid non-identical overlapping ranges

# Need to use split-file in order for `sed` calls below to work correctly
# RUN: split-file %s %t
# RUN: yaml2obj %t/test.yaml | llvm-dwarfdump --error-display=details --verify - | FileCheck %s
# CHECK: No errors.

# RUN: sed '0,/HighOffset: 0x6000/{s//HighOffset: 0x5999/}' %t/test.yaml | yaml2obj | not llvm-dwarfdump --error-display=details --verify - | FileCheck %s --check-prefix=CHECK-RANGES
# CHECK-RANGES: error: DIEs have overlapping address ranges

# RUN: sed '0,/Value: 0x77/{s/Value: 0x77/Value: 0x66/}' %t/test.yaml | yaml2obj | not llvm-dwarfdump --error-display=details --verify - | FileCheck %s --check-prefix=CHECK-HIGH-PC
# CHECK-HIGH-PC: error: DIEs have overlapping address ranges

# RUN: sed '0,/LowOffset: 0x880111/{s//LowOffset: 0x880112/}' %t/test.yaml | yaml2obj | not llvm-dwarfdump --error-display=details --verify - | FileCheck %s --check-prefix=CHECK-LEX-BLOCK
# CHECK-LEX-BLOCK: DIE has overlapping ranges in DW_AT_ranges attribute

#--- test.yaml
--- !ELF
FileHeader:
Class: ELFCLASS64
Data: ELFDATA2LSB
Type: ET_REL
Machine: EM_X86_64
DWARF:
debug_abbrev:
- Table:
- Tag: DW_TAG_compile_unit
Children: DW_CHILDREN_yes
Attributes:
- Attribute: DW_AT_producer
Form: DW_FORM_string
- Attribute: DW_AT_language
Form: DW_FORM_data2
- Attribute: DW_AT_name
Form: DW_FORM_string
- Attribute: DW_AT_low_pc
Form: DW_FORM_addr
- Attribute: DW_AT_high_pc
Form: DW_FORM_data8
- Tag: DW_TAG_subprogram
Children: DW_CHILDREN_no
Attributes:
- Attribute: DW_AT_name
Form: DW_FORM_string
- Attribute: DW_AT_low_pc
Form: DW_FORM_addr
- Attribute: DW_AT_high_pc
Form: DW_FORM_data8
- Tag: DW_TAG_subprogram
Children: DW_CHILDREN_no
Attributes:
- Attribute: DW_AT_name
Form: DW_FORM_string
- Attribute: DW_AT_ranges
Form: DW_FORM_sec_offset
- Tag: DW_TAG_base_type
Children: DW_CHILDREN_no
Attributes:
- Attribute: DW_AT_name
Form: DW_FORM_string
- Tag: DW_TAG_lexical_block
Children: DW_CHILDREN_no
Attributes:
- Attribute: DW_AT_ranges
Form: DW_FORM_sec_offset
debug_ranges:
- Offset: 0x0
AddrSize: 0x8
Entries:
- LowOffset: 0x1000
HighOffset: 0x2000
- LowOffset: 0x3000
HighOffset: 0x4000
- LowOffset: 0x5000 # Overlaps with 2nd range below
HighOffset: 0x6000
- LowOffset: 0x0
HighOffset: 0x0
- Offset: 0x50
AddrSize: 0x8
Entries:
- LowOffset: 0x2500
HighOffset: 0x2800
- LowOffset: 0x5000 # Overlaps with 3rd range above
HighOffset: 0x6000
- LowOffset: 0x7000
HighOffset: 0x8000
- LowOffset: 0x0
HighOffset: 0x0
- Offset: 0xA0 # Added Range List #3 for lexical block
AddrSize: 0x8
Entries:
- LowOffset: 0x880111
HighOffset: 0x881222
- LowOffset: 0x882333
HighOffset: 0x883444
- LowOffset: 0x880111 # Overlaps with 1st range in the same list
HighOffset: 0x881222
- LowOffset: 0x0 # End of list
HighOffset: 0x0
debug_info:
- Version: 4
Entries:
- AbbrCode: 1
Values:
- CStr: by_hand
- Value: 0x04
- CStr: CU1
- Value: 0x1000
- Value: 0x100
- AbbrCode: 4
Values:
- CStr: int
- AbbrCode: 2
Values:
- CStr: foo1
- Value: 0x1000
- Value: 0x10
- AbbrCode: 2
Values:
- CStr: foo2
- Value: 0x0 # Overlaps with 'foo3' below
- Value: 0x77
- AbbrCode: 2
Values:
- CStr: foo3
- Value: 0x0 # Overlaps with 'foo2' above
- Value: 0x77
- AbbrCode: 3
Values:
- CStr: func1_with_ranges
- Value: 0x0
- AbbrCode: 3
Values:
- CStr: func2_with_ranges
- Value: 0x50
- AbbrCode: 5 # Added lexical block using ranges
Values:
- Value: 0xA0 # Range list index in debug_ranges
- AbbrCode: 0
...
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,7 @@
# DW_AT_high_pc (0x0000000000000000)
#
# 0x00000033: DW_TAG_lexical_block
# DW_AT_low_pc (0x0000000000001000)
# DW_AT_low_pc (0x0000000000001001)
# DW_AT_high_pc (0x0000000000002000)
#
# 0x00000044: DW_TAG_lexical_block
Expand Down Expand Up @@ -47,7 +47,7 @@

# CHECK: error: DIEs have overlapping address ranges:
# CHECK: 0x00000044: DW_TAG_lexical_block
# CHECK: DW_AT_low_pc [DW_FORM_addr] (0x0000000000001000)
# CHECK: DW_AT_low_pc [DW_FORM_addr] (0x0000000000001001)
# CHECK: DW_AT_high_pc [DW_FORM_addr] (0x0000000000002000)

# CHECK: 0x00000033: DW_TAG_lexical_block
Expand All @@ -61,7 +61,7 @@
# CHECK: DW_AT_high_pc [DW_FORM_addr] (0x0000000000000000)

# CHECK: 0x00000044: DW_TAG_lexical_block
# CHECK: DW_AT_low_pc [DW_FORM_addr] (0x0000000000001000)
# CHECK: DW_AT_low_pc [DW_FORM_addr] (0x0000000000001001)
# CHECK: DW_AT_high_pc [DW_FORM_addr] (0x0000000000002000)


Expand Down Expand Up @@ -229,7 +229,7 @@ DWARF:
- Value: 0x0000000000002000
- AbbrCode: 0x00000003
Values:
- Value: 0x0000000000001000
- Value: 0x0000000000001001
- Value: 0x0000000000002000
- AbbrCode: 0x00000000
- AbbrCode: 0x00000000
Expand Down
2 changes: 1 addition & 1 deletion llvm/test/tools/llvm-dwarfutil/ELF/X86/verify.test
Original file line number Diff line number Diff line change
Expand Up @@ -150,7 +150,7 @@ DWARF:
Values:
- CStr: foo3
- Value: 0x0
- Value: 0x100
- Value: 0x80
- Value: 0x00000040
- AbbrCode: 0
...

0 comments on commit ad32576

Please sign in to comment.