Alexander Yermolovich a224c5199b [LLD][LLVM] CG Graph profile using relocations
Currently when .llvm.call-graph-profile is created by llvm it explicitly encodes the symbol indices. This section is basically a black box for post processing tools. For example, if we run strip -s on the object files the symbol table changes, but indices in that section do not. In non-visible behavior indices point to wrong symbols. The visible behavior indices point outside of Symbol table: "invalid symbol index".

This patch changes the format by using R_*_NONE relocations to indicate the from/to symbols. The Frequency (Weight) will still be in the .llvm.call-graph-profile, but symbol information will be in relocation section. In LLD information from both sections is used to reconstruct call graph profile. Relocations themselves will never be applied.

With this approach post processing tools that handle relocations correctly work for this section also. Tools can add/remove symbols and as long as they handle relocation sections with this approach information stays correct.

Doing a quick experiment with clang-13.
The size went up from 107KB to 322KB, aggregate of all the input sections. Size of clang-13 binary is ~118MB. For users of -fprofile-use/-fprofile-sample-use the size of object files will go up slightly, it will not impact final binary size.

Reviewed By: jhenderson, MaskRay

Differential Revision: https://reviews.llvm.org/D104080
2021-06-24 09:09:33 -07:00

238 lines
8.1 KiB
Plaintext

## Show that llvm-readelf + llvm-readobj demangle symbol names in symbol tables
## (including dynamic symbols), relocations (including dynamic relocations), and groups.
# RUN: yaml2obj %s -o %t.so
## Check LLVM output style.
# RUN: llvm-readobj --symbols --relocations --dyn-symbols --dyn-relocations \
# RUN: --elf-section-groups --cg-profile --addrsig \
# RUN: --demangle %t.so > %t.llvm.long
# RUN: llvm-readobj --symbols --relocations --dyn-symbols --dyn-relocations \
# RUN: --elf-section-groups --cg-profile --addrsig \
# RUN: -C %t.so > %t.llvm.short
# RUN: FileCheck %s --input-file %t.llvm.long --check-prefixes=LLVM-COMMON,LLVM-DEMANGLE
# RUN: diff %t.llvm.long %t.llvm.short
## Check that default is no demangling.
# RUN: llvm-readobj --symbols --relocations --dyn-symbols --dyn-relocations \
# RUN: --elf-section-groups --cg-profile --addrsig \
# RUN: %t.so > %t.llvm.default
# RUN: llvm-readobj --symbols --relocations --dyn-symbols --dyn-relocations \
# RUN: --elf-section-groups --cg-profile --addrsig \
# RUN: --demangle=false %t.so > %t.llvm.nodemangle
# RUN: FileCheck %s --input-file %t.llvm.default --check-prefixes=LLVM-COMMON,LLVM-MANGLED
# RUN: diff %t.llvm.default %t.llvm.nodemangle
# LLVM-COMMON: Relocations [
# LLVM-COMMON: Section {{.*}} .rela.text.foo {
# LLVM-DEMANGLE-NEXT: {{ }}foo(char){{ }}
# LLVM-MANGLED-NEXT: {{ }}_Z3fooc{{ }}
# LLVM-COMMON-NEXT: }
# LLVM-COMMON: ]
# LLVM-COMMON: Dynamic Relocations {
# LLVM-DEMANGLE-NEXT: {{ }}foo(int){{ }}
# LLVM-MANGLED-NEXT: {{ }}_Z3fooi{{ }}
# LLVM-COMMON-NEXT: }
# LLVM-COMMON: Symbols [
# LLVM-DEMANGLE: Name: foo(char){{ }}
# LLVM-DEMANGLE: Name: blah(float){{ }}
# LLVM-MANGLED: Name: _Z3fooc{{ }}
# LLVM-MANGLED: Name: _Z4blahf{{ }}
# LLVM-COMMON: ]
# LLVM-COMMON: DynamicSymbols [
# LLVM-DEMANGLE: Name: foo(int){{ }}
# LLVM-MANGLED: Name: _Z3fooi{{ }}
# LLVM-COMMON: ]
# LLVM-COMMON: Groups {
# LLVM-DEMANGLE: Signature: foo(char){{$}}
# LLVM-MANGLED: Signature: _Z3fooc{{$}}
# LLVM-COMMON: }
# LLVM-COMMON: CGProfile [
# LLVM-DEMANGLE: From: foo(char){{ }}
# LLVM-DEMANGLE: To: blah(float){{ }}
# LLVM-MANGLED: From: _Z3fooc{{ }}
# LLVM-MANGLED: To: _Z4blahf{{ }}
# LLVM-COMMON: ]
# LLVM-COMMON: Addrsig [
# LLVM-DEMANGLE-NEXT: Sym: foo(char){{ }}
# LLVM-DEMANGLE-NEXT: Sym: blah(float){{ }}
# LLVM-MANGLED-NEXT: Sym: _Z3fooc{{ }}
# LLVM-MANGLED-NEXT: Sym: _Z4blahf{{ }}
# LLVM-COMMON-NEXT: ]
## Check GNU output style.
# RUN: llvm-readelf --symbols --relocations --dyn-symbols --dyn-relocations \
# RUN: --elf-section-groups --addrsig --demangle %t.so > %t.gnu.long
# RUN: llvm-readelf --symbols --relocations --dyn-symbols --dyn-relocations \
# RUN: --elf-section-groups --addrsig -C %t.so > %t.gnu.short
# RUN: FileCheck %s --input-file %t.gnu.long --check-prefixes=GNU-COMMON,GNU-DEMANGLE
# RUN: diff %t.gnu.long %t.gnu.short
## Check that default is no demangling.
# RUN: llvm-readelf --symbols --relocations --dyn-symbols --dyn-relocations \
# RUN: --elf-section-groups --addrsig %t.so > %t.gnu.default
# RUN: llvm-readelf --symbols --relocations --dyn-symbols --dyn-relocations \
# RUN: --elf-section-groups --addrsig --demangle=false %t.so > %t.gnu.nodemangle
# RUN: FileCheck %s --input-file %t.gnu.default --check-prefixes=GNU-COMMON,GNU-MANGLED
# RUN: diff %t.gnu.default %t.gnu.nodemangle
# GNU-COMMON: Relocation section '.rela.text.foo' at offset {{.*}} contains 1 entries:
# GNU-COMMON-NEXT: Offset Info Type Symbol's Value Symbol's Name + Addend
# GNU-DEMANGLE-NEXT: foo(char){{ }}
# GNU-MANGLED-NEXT: _Z3fooc{{ }}
# GNU-COMMON: 'RELA' relocation section at offset {{.*}} contains 24 bytes:
# GNU-COMMON-NEXT: Offset Info Type Symbol's Value Symbol's Name + Addend
# GNU-DEMANGLE-NEXT: foo(int){{ }}
# GNU-MANGLED-NEXT: _Z3fooi{{ }}
# GNU-COMMON: Symbol table '.dynsym' contains 2 entries:
# GNU-COMMON-NEXT: Num: Value Size Type Bind Vis Ndx Name
# GNU-COMMON-NEXT: 0: 0000000000000000 0 NOTYPE LOCAL DEFAULT UND
# GNU-DEMANGLE-NEXT: foo(int){{$}}
# GNU-MANGLED-NEXT: _Z3fooi{{$}}
# GNU-COMMON: Symbol table '.symtab' contains 3 entries:
# GNU-COMMON-NEXT: Num: Value Size Type Bind Vis Ndx Name
# GNU-COMMON-NEXT: 0: 0000000000000000 0 NOTYPE LOCAL DEFAULT UND
# GNU-DEMANGLE-NEXT: foo(char){{$}}
# GNU-DEMANGLE-NEXT: blah(float){{$}}
# GNU-MANGLED-NEXT: _Z3fooc{{$}}
# GNU-MANGLED-NEXT: _Z4blahf{{$}}
# GNU-COMMON: COMDAT group section [{{.*}}] `.group'
# GNU-DEMANGLE-SAME: [foo(char)]
# GNU-MANGLED-SAME: [_Z3fooc]
# GNU-COMMON: Address-significant symbols section '.llvm_addrsig' contains 2 entries:
# GNU-COMMON: Num: Name
# GNU-DEMANGLE-NEXT: 1: foo(char)
# GNU-DEMANGLE-NEXT: 2: blah(float)
# GNU-MANGLED-NEXT: 1: _Z3fooc
# GNU-MANGLED-NEXT: 2: _Z4blahf
!ELF
FileHeader:
Class: ELFCLASS64
Data: ELFDATA2LSB
Type: ET_DYN
Machine: EM_X86_64
Sections:
- Name: .dynstr
Type: SHT_STRTAB
Flags: [ SHF_ALLOC ]
AddressAlign: 0x100
EntSize: 0x1
- Name: .dynsym
Type: SHT_DYNSYM
Flags: [ SHF_ALLOC ]
Link: .dynstr
Address: 0x100
AddressAlign: 0x100
EntSize: 0x18
- Name: .rela.dyn
Type: SHT_RELA
Flags: [ SHF_ALLOC ]
Link: .dynsym
Info: .text.foo
Address: 0x200
AddressAlign: 0x100
EntSize: 0x18
Relocations:
- Offset: 0x10
Symbol: _Z3fooi
Type: R_X86_64_PC32
Addend: 0x4
- Name: .dynamic
Type: SHT_DYNAMIC
Flags: [ SHF_ALLOC ]
Address: 0x0000000000001000
Link: .dynstr
AddressAlign: 0x0000000000001000
EntSize: 0x0000000000000010
Entries:
- Tag: DT_STRTAB
Value: 0x0000000000000000
- Tag: DT_STRSZ
Value: 0x0000000000000009
- Tag: DT_SYMTAB
Value: 0x0000000000000100
- Tag: DT_SYMENT
Value: 0x0000000000000018
- Tag: DT_RELA
Value: 0x0000000000000200
- Tag: DT_RELASZ
Value: 0x0000000000000018
- Tag: DT_RELAENT
Value: 0x0000000000000018
- Tag: DT_NULL
Value: 0x0000000000000000
- Name: .text.foo
Type: SHT_PROGBITS
Flags: [ SHF_ALLOC, SHF_EXECINSTR, SHF_GROUP ]
Size: 0x40
Address: 0x2000
AddressAlign: 0x2000
- Name: .group
Type: SHT_GROUP
Link: .symtab
Info: _Z3fooc
Members:
- SectionOrType: GRP_COMDAT
- SectionOrType: .text.foo
- Name: .rela.text.foo
Type: SHT_RELA
Link: .symtab
Info: .text.foo
Relocations:
- Offset: 0x10
Symbol: _Z3fooc
Type: R_X86_64_PC32
- Name: .llvm.call-graph-profile
Type: SHT_LLVM_CALL_GRAPH_PROFILE
Link: .symtab
EntSize: 8
Content: "2000000000000000"
- Name: .rela.llvm.call-graph-profile
Type: SHT_RELA
Info: .llvm.call-graph-profile
Relocations:
- Symbol: 1
Type: R_X86_64_NONE
- Offset: 0x1
Symbol: 2
Type: R_X86_64_NONE
- Name: .llvm_addrsig
Type: SHT_LLVM_ADDRSIG
Link: .symtab
Content: "0102"
Symbols:
- Name: _Z3fooc
Type: STT_FUNC
Section: .text.foo
Binding: STB_GLOBAL
- Name: _Z4blahf
Type: STT_FUNC
Section: .text.foo
Binding: STB_GLOBAL
DynamicSymbols:
- Name: _Z3fooi
Binding: STB_GLOBAL
ProgramHeaders:
- Type: PT_LOAD
Flags: [ PF_R, PF_X ]
VAddr: 0x0
FirstSec: .dynstr
LastSec: .text.foo
- Type: PT_DYNAMIC
Flags: [ PF_R ]
VAddr: 0x1000
FirstSec: .dynamic
LastSec: .dynamic