[llvm] [BOLT] Fix relocations handling (PR #100890)
via llvm-commits
llvm-commits at lists.llvm.org
Sat Jul 27 12:19:08 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-bolt
Author: Vladislav Khmelevsky (yota9)
<details>
<summary>Changes</summary>
After porting BOLT to RISCV some of the relocations were broken on both
AArch64 and X86.
On AArch64 the example of broken relocations would be GOT, during
handling them, we should replace the symbol to __BOLT_got_zero in order
to address GOT entry, not the symbol that addresses this entry. This is
done further in code, so it is too early to add rel here.
On X86 it is a mistake to add relocations without addend. This is the
exact problem that is raised on #<!-- -->97937. Due to different code generation
I had to use gcc-generated yaml test, since with clang I wasn't able to
reproduce problem.
Added tests for both architectures and made the problematic condition
riscV-specific.
---
Full diff: https://github.com/llvm/llvm-project/pull/100890.diff
5 Files Affected:
- (modified) bolt/lib/Rewrite/RewriteInstance.cpp (+1-1)
- (added) bolt/test/AArch64/Inputs/build_id.ldscript (+9)
- (added) bolt/test/AArch64/build_id.c (+28)
- (added) bolt/test/X86/Inputs/build_id.yaml (+326)
- (added) bolt/test/X86/build_id.test (+8)
``````````diff
diff --git a/bolt/lib/Rewrite/RewriteInstance.cpp b/bolt/lib/Rewrite/RewriteInstance.cpp
index 33ebae3b6e6de..051d7e11c6654 100644
--- a/bolt/lib/Rewrite/RewriteInstance.cpp
+++ b/bolt/lib/Rewrite/RewriteInstance.cpp
@@ -2603,7 +2603,7 @@ void RewriteInstance::handleRelocation(const SectionRef &RelocatedSection,
Expected<StringRef> SectionName = Section->getName();
if (SectionName && !SectionName->empty())
ReferencedSection = BC->getUniqueSectionByName(*SectionName);
- } else if (ReferencedSymbol && ContainingBF &&
+ } else if (BC->isRISCV() && ReferencedSymbol && ContainingBF &&
(cantFail(Symbol.getFlags()) & SymbolRef::SF_Absolute)) {
// This might be a relocation for an ABS symbols like __global_pointer$ on
// RISC-V
diff --git a/bolt/test/AArch64/Inputs/build_id.ldscript b/bolt/test/AArch64/Inputs/build_id.ldscript
new file mode 100644
index 0000000000000..0af8e960f491b
--- /dev/null
+++ b/bolt/test/AArch64/Inputs/build_id.ldscript
@@ -0,0 +1,9 @@
+SECTIONS
+{
+ PROVIDE (__executable_start = SEGMENT_START("text-segment", 0x400000)); . = SEGMENT_START("text-segment", 0x400000) + SIZEOF_HEADERS;
+ .note.gnu.build-id (0x400400):
+ {
+ build_id_note = ABSOLUTE(.);
+ *(.note.gnu.build-id)
+ }
+}
diff --git a/bolt/test/AArch64/build_id.c b/bolt/test/AArch64/build_id.c
new file mode 100644
index 0000000000000..bc7832afe5304
--- /dev/null
+++ b/bolt/test/AArch64/build_id.c
@@ -0,0 +1,28 @@
+// This test checks that referencing build_id through GOT table
+// would result in GOT access after disassembly, not directly
+// to build_id address.
+
+// RUN: %clang %cflags -fuse-ld=lld -Wl,-T,%S/Inputs/build_id.ldscript -Wl,-q \
+// RUN: -Wl,--no-relax -Wl,--build-id=sha1 %s -o %t.exe
+// RUN: llvm-bolt -print-disasm --print-only=get_build_id %t.exe -o %t.bolt | \
+// RUN: FileCheck %s
+
+// CHECK: adrp [[REG:x[0-28]+]], __BOLT_got_zero
+// CHECK: ldr x{{.*}}, [[[REG]], :lo12:__BOLT_got_zero{{.*}}]
+
+struct build_id_note {
+ char pad[16];
+ char hash[20];
+};
+
+extern const struct build_id_note build_id_note;
+
+__attribute__((noinline))
+char get_build_id() {
+ return build_id_note.hash[0];
+}
+
+int main() {
+ get_build_id();
+ return 0;
+}
diff --git a/bolt/test/X86/Inputs/build_id.yaml b/bolt/test/X86/Inputs/build_id.yaml
new file mode 100644
index 0000000000000..af012904ff950
--- /dev/null
+++ b/bolt/test/X86/Inputs/build_id.yaml
@@ -0,0 +1,326 @@
+--- !ELF
+FileHeader:
+ Class: ELFCLASS64
+ Data: ELFDATA2LSB
+ Type: ET_EXEC
+ Machine: EM_X86_64
+ Entry: 0x4010A0
+ProgramHeaders:
+ - Type: PT_PHDR
+ Flags: [ PF_R ]
+ VAddr: 0x400040
+ Align: 0x8
+ Offset: 0x40
+ - Type: PT_INTERP
+ Flags: [ PF_R ]
+ FirstSec: .interp
+ LastSec: .interp
+ VAddr: 0x400444
+ Offset: 0x444
+ - Type: PT_LOAD
+ Flags: [ PF_X, PF_R ]
+ FirstSec: .init
+ LastSec: .fini
+ VAddr: 0x401000
+ Align: 0x1000
+ Offset: 0x1000
+ - Type: PT_LOAD
+ Flags: [ PF_R ]
+ FirstSec: .rodata
+ LastSec: .rodata
+ VAddr: 0x402000
+ Align: 0x1000
+ Offset: 0x2000
+ - Type: PT_LOAD
+ Flags: [ PF_W, PF_R ]
+ FirstSec: .init_array
+ LastSec: .bss
+ VAddr: 0x403DD8
+ Align: 0x1000
+ Offset: 0x2DD8
+ - Type: PT_DYNAMIC
+ Flags: [ PF_W, PF_R ]
+ FirstSec: .dynamic
+ LastSec: .dynamic
+ VAddr: 0x403DE8
+ Align: 0x8
+ Offset: 0x2DE8
+ - Type: PT_NOTE
+ Flags: [ PF_R ]
+ FirstSec: .note.gnu.build-id
+ LastSec: .note.ABI-tag
+ VAddr: 0x400400
+ Align: 0x4
+ Offset: 0x400
+Sections:
+ - Name: .note.gnu.build-id
+ Type: SHT_NOTE
+ Flags: [ SHF_ALLOC ]
+ Address: 0x400400
+ AddressAlign: 0x4
+ Offset: 0x400
+ Notes:
+ - Name: GNU
+ Desc: 3C34F7D1612996940C48F98DC272543BC3C9C956
+ Type: NT_PRPSINFO
+ - Name: .note.ABI-tag
+ Type: SHT_NOTE
+ Flags: [ SHF_ALLOC ]
+ Address: 0x400424
+ AddressAlign: 0x4
+ Notes:
+ - Name: GNU
+ Desc: '00000000030000000200000000000000'
+ Type: NT_VERSION
+ - Name: .interp
+ Type: SHT_PROGBITS
+ Flags: [ SHF_ALLOC ]
+ Address: 0x400444
+ AddressAlign: 0x1
+ Content: 2F6C696236342F6C642D6C696E75782D7838362D36342E736F2E3200
+ - Name: .gnu.hash
+ Type: SHT_GNU_HASH
+ Flags: [ SHF_ALLOC ]
+ Address: 0x400460
+ Link: .dynsym
+ AddressAlign: 0x8
+ Header:
+ SymNdx: 0x7
+ Shift2: 0x6
+ BloomFilter: [ 0x810000 ]
+ HashBuckets: [ 0x7, 0x0 ]
+ HashValues: [ 0x6DCE65D1 ]
+ - Name: .dynsym
+ Type: SHT_DYNSYM
+ Flags: [ SHF_ALLOC ]
+ Address: 0x400488
+ Link: .dynstr
+ AddressAlign: 0x8
+ - Name: .dynstr
+ Type: SHT_STRTAB
+ Flags: [ SHF_ALLOC ]
+ Address: 0x400548
+ AddressAlign: 0x1
+ - Name: .gnu.version
+ Type: SHT_GNU_versym
+ Flags: [ SHF_ALLOC ]
+ Address: 0x4005F2
+ Link: .dynsym
+ AddressAlign: 0x2
+ Entries: [ 0, 2, 3, 1, 1, 4, 1, 2 ]
+ - Name: .gnu.version_r
+ Type: SHT_GNU_verneed
+ Flags: [ SHF_ALLOC ]
+ Address: 0x400608
+ Link: .dynstr
+ AddressAlign: 0x8
+ Dependencies:
+ - Version: 1
+ File: libc.so.6
+ Entries:
+ - Name: GLIBC_2.3.4
+ Hash: 157882740
+ Flags: 0
+ Other: 4
+ - Name: GLIBC_2.34
+ Hash: 110530996
+ Flags: 0
+ Other: 3
+ - Name: GLIBC_2.2.5
+ Hash: 157882997
+ Flags: 0
+ Other: 2
+ - Name: .init
+ Type: SHT_PROGBITS
+ Flags: [ SHF_ALLOC, SHF_EXECINSTR ]
+ Address: 0x401000
+ AddressAlign: 0x4
+ Offset: 0x1000
+ Content: F30F1EFA4883EC08488B05D92F00004885C07402FFD04883C408C3
+ - Name: .plt.sec
+ Type: SHT_PROGBITS
+ Flags: [ SHF_ALLOC, SHF_EXECINSTR ]
+ Address: 0x401060
+ AddressAlign: 0x10
+ EntSize: 0x10
+ Content: F30F1EFAF2FF25AD2F00000F1F440000F30F1EFAF2FF25A52F00000F1F440000
+ - Name: .text
+ Type: SHT_PROGBITS
+ Flags: [ SHF_ALLOC, SHF_EXECINSTR ]
+ Address: 0x401080
+ AddressAlign: 0x10
+ Content: F30F1EFA4883EC0831C0E80101000031C04883C408C3662E0F1F840000000000F30F1EFA31ED4989D15E4889E24883E4F050544531C031C9488D3DC1FFFFFFFF15132F0000F4662E0F1F840000000000488D3D612F0000488D055A2F00004839F87415488B05F62E00004885C07409FFE00F1F8000000000C30F1F8000000000488D3D312F0000488D352A2F00004829FE4889F048C1EE3F48C1F8034801C648D1FE7414488B05C52E00004885C07408FFE0660F1F440000C30F1F8000000000F30F1EFA803DED2E000000752B5548833DA22E0000004889E5740C488B3DCE2E0000E8E9FEFFFFE864FFFFFFC605C52E0000015DC30F1F00C30F1F8000000000F30F1EFAE977FFFFFF0F1F8000000000F30F1EFA415455488D2D660E000053488D1D6AF2FFFF4C8D6314660F1F4400000FB6134889EEBF0100000031C04883C301E8AAFEFFFF4C39E375E55BBF0A0000005D415CE987FEFFFF
+ - Name: .fini
+ Type: SHT_PROGBITS
+ Flags: [ SHF_ALLOC, SHF_EXECINSTR ]
+ Address: 0x4011DC
+ AddressAlign: 0x4
+ Content: F30F1EFA4883EC084883C408C3
+ - Name: .rodata
+ Type: SHT_PROGBITS
+ Flags: [ SHF_ALLOC ]
+ Address: 0x402000
+ AddressAlign: 0x4
+ Offset: 0x2000
+ Content: '0100020025303268687800'
+ - Name: .init_array
+ Type: SHT_INIT_ARRAY
+ Flags: [ SHF_WRITE, SHF_ALLOC ]
+ Address: 0x403DD8
+ AddressAlign: 0x8
+ EntSize: 0x8
+ Offset: 0x2DD8
+ Content: '8011400000000000'
+ - Name: .fini_array
+ Type: SHT_FINI_ARRAY
+ Flags: [ SHF_WRITE, SHF_ALLOC ]
+ Address: 0x403DE0
+ AddressAlign: 0x8
+ EntSize: 0x8
+ Content: '4011400000000000'
+ - Name: .dynamic
+ Type: SHT_DYNAMIC
+ Flags: [ SHF_WRITE, SHF_ALLOC ]
+ Address: 0x403DE8
+ Link: .dynstr
+ AddressAlign: 0x8
+ Entries:
+ - Tag: DT_NEEDED
+ Value: 0x37
+ - Tag: DT_INIT
+ Value: 0x401000
+ - Tag: DT_FINI
+ Value: 0x4011DC
+ - Tag: DT_INIT_ARRAY
+ Value: 0x403DD8
+ - Tag: DT_INIT_ARRAYSZ
+ Value: 0x8
+ - Tag: DT_FINI_ARRAY
+ Value: 0x403DE0
+ - Tag: DT_FINI_ARRAYSZ
+ Value: 0x8
+ - Tag: DT_GNU_HASH
+ Value: 0x400460
+ - Tag: DT_STRTAB
+ Value: 0x400548
+ - Tag: DT_SYMTAB
+ Value: 0x400488
+ - Tag: DT_STRSZ
+ Value: 0xA9
+ - Tag: DT_SYMENT
+ Value: 0x18
+ - Tag: DT_DEBUG
+ Value: 0x0
+ - Tag: DT_PLTGOT
+ Value: 0x404000
+ - Tag: DT_PLTRELSZ
+ Value: 0x30
+ - Tag: DT_PLTREL
+ Value: 0x7
+ - Tag: DT_FLAGS
+ Value: 0x8
+ - Tag: DT_FLAGS_1
+ Value: 0x8000001
+ - Tag: DT_VERNEED
+ Value: 0x400608
+ - Tag: DT_VERNEEDNUM
+ Value: 0x1
+ - Tag: DT_VERSYM
+ Value: 0x4005F2
+ - Tag: DT_RELACOUNT
+ Value: 0x3
+ - Tag: DT_NULL
+ Value: 0x0
+ - Name: .data
+ Type: SHT_PROGBITS
+ Flags: [ SHF_WRITE, SHF_ALLOC ]
+ Address: 0x404028
+ AddressAlign: 0x8
+ Content: '00000000000000003040400000000000'
+ - Name: .tm_clone_table
+ Type: SHT_PROGBITS
+ Flags: [ SHF_WRITE, SHF_ALLOC ]
+ Address: 0x404038
+ AddressAlign: 0x8
+ - Name: .bss
+ Type: SHT_NOBITS
+ Flags: [ SHF_WRITE, SHF_ALLOC ]
+ Address: 0x404038
+ AddressAlign: 0x1
+ Size: 0x8
+ - Name: .rela.text
+ Type: SHT_RELA
+ Flags: [ SHF_INFO_LINK ]
+ Link: .symtab
+ AddressAlign: 0x8
+ Info: .text
+ Relocations:
+ - Offset: 0x40108B
+ Symbol: print_build_id
+ Type: R_X86_64_PLT32
+ Addend: -4
+ - Offset: 0x4010BB
+ Symbol: main
+ Type: R_X86_64_PC32
+ Addend: -4
+ - Offset: 0x4011A2
+ Symbol: build_id_note
+ Type: R_X86_64_PC32
+ Addend: 12
+ - Type: SectionHeaderTable
+ Sections:
+ - Name: .note.gnu.build-id
+ - Name: .note.ABI-tag
+ - Name: .interp
+ - Name: .gnu.hash
+ - Name: .dynsym
+ - Name: .dynstr
+ - Name: .gnu.version
+ - Name: .gnu.version_r
+ - Name: .init
+ - Name: .plt.sec
+ - Name: .text
+ - Name: .rela.text
+ - Name: .fini
+ - Name: .rodata
+ - Name: .init_array
+ - Name: .fini_array
+ - Name: .dynamic
+ - Name: .data
+ - Name: .tm_clone_table
+ - Name: .bss
+ - Name: .symtab
+ - Name: .strtab
+ - Name: .shstrtab
+Symbols:
+ - Name: print_build_id
+ Type: STT_FUNC
+ Section: .text
+ Binding: STB_GLOBAL
+ Value: 0x401190
+ Size: 0x49
+ - Name: _end
+ Section: .bss
+ Binding: STB_GLOBAL
+ Value: 0x404040
+ - Name: _start
+ Type: STT_FUNC
+ Section: .text
+ Binding: STB_GLOBAL
+ Value: 0x4010A0
+ Size: 0x26
+ - Name: __bss_start
+ Section: .bss
+ Binding: STB_GLOBAL
+ Value: 0x404038
+ - Name: main
+ Type: STT_FUNC
+ Section: .text
+ Binding: STB_GLOBAL
+ Value: 0x401080
+ Size: 0x16
+ - Name: build_id_note
+ Index: SHN_ABS
+ Binding: STB_GLOBAL
+ Value: 0x400400
+...
diff --git a/bolt/test/X86/build_id.test b/bolt/test/X86/build_id.test
new file mode 100644
index 0000000000000..8d28e12dbf94f
--- /dev/null
+++ b/bolt/test/X86/build_id.test
@@ -0,0 +1,8 @@
+// This test checks that relocation addend used to address build_id fields
+// is properly disassembled by BOLT.
+
+RUN: yaml2obj %p/Inputs/build_id.yaml &> %t.exe
+RUN: llvm-bolt -print-disasm --print-only=print_build_id %t.exe -o %t.bolt | \
+RUN: FileCheck %s
+
+CHECK: leaq build_id_note+16(%rip), %rbx
``````````
</details>
https://github.com/llvm/llvm-project/pull/100890
More information about the llvm-commits
mailing list