[ELF] --emit-relocs: fix st_value of STT_SECTION in the presence of a gap before the first input section

In the presence of a gap, the st_value field of a STT_SECTION symbol is the
address of the first input section (incorrect if there is a gap). Set it to the
output section address instead.

In -r mode, this bug can cause an incorrect non-zero st_value of a STT_SECTION
symbol (while output sections have zero addresses, input sections may have
non-zero outSecOff).  The non-zero st_value can cause the final link to have
incorrect relocation computation (both GNU ld and LLD add st_value of the
STT_SECTION symbol to the output section address).

Reviewed By: grimar

Differential Revision: https://reviews.llvm.org/D90520
This commit is contained in:
Fangrui Song
2020-11-02 08:37:15 -08:00
parent 2ca4406de8
commit 2fc704a0a5
3 changed files with 57 additions and 4 deletions

View File

@@ -2198,9 +2198,8 @@ template <class ELFT> void SymbolTableSection<ELFT>::writeTo(uint8_t *buf) {
else
eSym->st_size = sym->getSize();
// st_value is usually an address of a symbol, but that has a
// special meaning for uninstantiated common symbols (this can
// occur if -r is given).
// st_value is usually an address of a symbol, but that has a special
// meaning for uninstantiated common symbols (--no-define-common).
if (BssSection *commonSec = getCommonSec(ent.sym))
eSym->st_value = commonSec->alignment;
else if (isDefinedHere)

View File

@@ -811,9 +811,12 @@ template <class ELFT> void Writer<ELFT>::addSectionSymbols() {
if (isa<SyntheticSection>(isec) && !(isec->flags & SHF_MERGE))
continue;
// Set the symbol to be relative to the output section so that its st_value
// equals the output section address. Note, there may be a gap between the
// start of the output section and isec.
auto *sym =
make<Defined>(isec->file, "", STB_LOCAL, /*stOther=*/0, STT_SECTION,
/*value=*/0, /*size=*/0, isec);
/*value=*/0, /*size=*/0, isec->getOutputSection());
in.symTab->addSymbol(sym);
}
}

View File

@@ -0,0 +1,51 @@
# REQUIRES: x86
## Test st_value of the STT_SECTION symbol equals the output section address,
## instead of the first input section address.
# RUN: split-file %s %t
# RUN: llvm-mc -filetype=obj -triple=x86_64 %t/asm -o %t.o
# RUN: ld.lld --emit-relocs -T %t/lds %t.o -o %t.out
# RUN: llvm-readelf -S -r -s %t.out | FileCheck %s --check-prefix=EXE
## In -r mode, section addresses are zeros, hence the st_value fields of
## STT_SECTION are zeros.
# RUN: ld.lld -r -T %t/lds %t.o -o %t.ro
# RUN: llvm-readelf -S -r -s %t.ro | FileCheck %s --check-prefix=RO
# EXE: [Nr] Name Type Address
# EXE-NEXT: [ 0]
# EXE-NEXT: [ 1] .text PROGBITS 0000000000000000
# EXE-NEXT: [ 2] .bss NOBITS 000000000000000a
# EXE: R_X86_64_64 {{.*}} .bss + 1
# EXE: Symbol table '.symtab' contains 4 entries:
# EXE-NEXT: Num: Value Size Type Bind Vis Ndx Name
# EXE-NEXT: 0: 0000000000000000 0 NOTYPE LOCAL DEFAULT UND
# EXE-NEXT: 1: 000000000000000a 0 SECTION LOCAL DEFAULT 2 .bss
# EXE-NEXT: 2: 0000000000000000 0 SECTION LOCAL DEFAULT 1 .text
# EXE-NEXT: 3: 0000000000000000 0 SECTION LOCAL DEFAULT 4 .comment
# RO: [Nr] Name Type Address
# RO-NEXT: [ 0]
# RO-NEXT: [ 1] .bss NOBITS 0000000000000000
# RO: R_X86_64_64 {{.*}} .bss + 1
# RO: Symbol table '.symtab' contains 3 entries:
# RO-NEXT: Num: Value Size Type Bind Vis Ndx Name
# RO-NEXT: 0: 0000000000000000 0 NOTYPE LOCAL DEFAULT UND
# RO-NEXT: 1: 0000000000000000 0 SECTION LOCAL DEFAULT 1 .bss
# RO-NEXT: 2: 0000000000000000 0 SECTION LOCAL DEFAULT 2 .text
#--- asm
movabsq .bss, %rax
.bss
.byte 0
#--- lds
SECTIONS {
.bss : { BYTE(0) *(.bss) }
}