Skip to content

Commit e196e9b

Browse files
jeremyd2019jeremyd2019
committed
[LLD][COFF] add __{data,bss}_{start,end}__ symbols for Cygwin support (llvm#136180)
Cygwin requires these symbols for its fork emulation to know what data to copy into the child. GNU ld defines these symbols for MinGW targets also, so do the same here. Cygwin also has the `.data_cygwin_nocopy` section, which is merged into `.data` outside the `__data_start__` to `__data_end__` range. This excludes it from fork's copying. AFAIK it's only used by the Cygwin DLL itself (which requires a custom linker script to link, that's not supported by LLD), but the section is included in GNU ld's default linker script so handle it here too. Signed-off-by: Jeremy Drake <[email protected]>
1 parent 72c2ecd commit e196e9b

File tree

3 files changed

+135
-2
lines changed

3 files changed

+135
-2
lines changed

‎lld/COFF/Driver.cpp

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2018,6 +2018,7 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
20182018
parseMerge(".ctors=.rdata");
20192019
parseMerge(".dtors=.rdata");
20202020
parseMerge(".CRT=.rdata");
2021+
parseMerge(".data_cygwin_nocopy=.data");
20212022
}
20222023

20232024
// Handle /section
@@ -2474,6 +2475,10 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
24742475
if (config->mingw) {
24752476
symtab.addAbsolute(symtab.mangle("__CTOR_LIST__"), 0);
24762477
symtab.addAbsolute(symtab.mangle("__DTOR_LIST__"), 0);
2478+
symtab.addAbsolute("__data_start__", 0);
2479+
symtab.addAbsolute("__data_end__", 0);
2480+
symtab.addAbsolute("__bss_start__", 0);
2481+
symtab.addAbsolute("__bss_end__", 0);
24772482
}
24782483
if (config->debug || config->buildIDHash != BuildIDHash::None)
24792484
if (symtab.findUnderscore("__buildid"))

‎lld/COFF/Writer.cpp

Lines changed: 31 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -240,6 +240,7 @@ class Writer {
240240
void createRuntimePseudoRelocs();
241241
void createECChunks();
242242
void insertCtorDtorSymbols();
243+
void insertBssDataStartEndSymbols();
243244
void markSymbolsWithRelocations(ObjFile *file, SymbolRVASet &usedSymbols);
244245
void createGuardCFTables();
245246
void markSymbolsForRVATable(ObjFile *file,
@@ -315,6 +316,7 @@ class Writer {
315316

316317
OutputSection *textSec;
317318
OutputSection *hexpthkSec;
319+
OutputSection *bssSec;
318320
OutputSection *rdataSec;
319321
OutputSection *buildidSec;
320322
OutputSection *dataSec;
@@ -1069,7 +1071,7 @@ void Writer::createSections() {
10691071
textSec = createSection(".text", code | r | x);
10701072
if (isArm64EC(ctx.config.machine))
10711073
hexpthkSec = createSection(".hexpthk", code | r | x);
1072-
createSection(".bss", bss | r | w);
1074+
bssSec = createSection(".bss", bss | r | w);
10731075
rdataSec = createSection(".rdata", data | r);
10741076
buildidSec = createSection(".buildid", data | r);
10751077
dataSec = createSection(".data", data | r | w);
@@ -1243,8 +1245,10 @@ void Writer::createMiscChunks() {
12431245
if (config->autoImport)
12441246
createRuntimePseudoRelocs();
12451247

1246-
if (config->mingw)
1248+
if (config->mingw) {
12471249
insertCtorDtorSymbols();
1250+
insertBssDataStartEndSymbols();
1251+
}
12481252
}
12491253

12501254
// Create .idata section for the DLL-imported symbol table.
@@ -2347,6 +2351,31 @@ void Writer::insertCtorDtorSymbols() {
23472351
dtorListHead);
23482352
}
23492353

2354+
// MinGW (really, Cygwin) specific.
2355+
// The Cygwin startup code uses __data_start__ __data_end__ __bss_start__
2356+
// and __bss_end__ to know what to copy during fork emulation.
2357+
void Writer::insertBssDataStartEndSymbols() {
2358+
if (!dataSec->chunks.empty()) {
2359+
Symbol *dataStartSym = ctx.symtab.find("__data_start__");
2360+
Symbol *dataEndSym = ctx.symtab.find("__data_end__");
2361+
Chunk *endChunk = dataSec->chunks.back();
2362+
replaceSymbol<DefinedSynthetic>(dataStartSym, dataStartSym->getName(),
2363+
dataSec->chunks.front());
2364+
replaceSymbol<DefinedSynthetic>(dataEndSym, dataEndSym->getName(), endChunk,
2365+
endChunk->getSize());
2366+
}
2367+
2368+
if (!bssSec->chunks.empty()) {
2369+
Symbol *bssStartSym = ctx.symtab.find("__bss_start__");
2370+
Symbol *bssEndSym = ctx.symtab.find("__bss_end__");
2371+
Chunk *endChunk = bssSec->chunks.back();
2372+
replaceSymbol<DefinedSynthetic>(bssStartSym, bssStartSym->getName(),
2373+
bssSec->chunks.front());
2374+
replaceSymbol<DefinedSynthetic>(bssEndSym, bssEndSym->getName(), endChunk,
2375+
endChunk->getSize());
2376+
}
2377+
}
2378+
23502379
// Handles /section options to allow users to overwrite
23512380
// section attributes.
23522381
void Writer::setSectionPermissions() {

‎lld/test/COFF/cygwin-symbols.s

Lines changed: 99 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,99 @@
1+
# REQUIRES: x86
2+
# RUN: split-file %s %t.dir && cd %t.dir
3+
4+
# RUN: llvm-mc -triple=x86_64-windows-cygnus -filetype=obj -o data-no-bss.obj data-no-bss.s
5+
# RUN: lld-link -lldmingw -entry:main data-no-bss.obj -out:data-no-bss.exe
6+
# RUN: llvm-objdump -s data-no-bss.exe | FileCheck --check-prefix=DATANOBSS %s
7+
8+
# RUN: llvm-mc -triple=x86_64-windows-cygnus -filetype=obj -o bss-no-data.obj bss-no-data.s
9+
# RUN: lld-link -lldmingw -entry:main bss-no-data.obj -out:bss-no-data.exe
10+
# RUN: llvm-objdump -s bss-no-data.exe | FileCheck --check-prefix=BSSNODATA %s
11+
12+
# RUN: llvm-mc -triple=x86_64-windows-cygnus -filetype=obj -o data-and-bss.obj data-and-bss.s
13+
# RUN: lld-link -lldmingw -entry:main data-and-bss.obj -out:data-and-bss.exe
14+
# RUN: llvm-objdump -s data-and-bss.exe | FileCheck --check-prefix=DATAANDBSS %s
15+
16+
#--- data-no-bss.s
17+
.globl main
18+
main:
19+
nop
20+
21+
.data
22+
.quad 1
23+
.byte 2
24+
25+
.section .data_cygwin_nocopy, "w"
26+
.align 4
27+
.quad 3
28+
.byte 4
29+
30+
.section .test, "w"
31+
.quad __data_start__
32+
.quad __data_end__
33+
.quad __bss_start__
34+
.quad __bss_end__
35+
36+
#--- bss-no-data.s
37+
.globl main
38+
main:
39+
nop
40+
41+
.bss
42+
.zero 8192
43+
44+
.section .test, "w"
45+
.quad __data_start__
46+
.quad __data_end__
47+
.quad __bss_start__
48+
.quad __bss_end__
49+
50+
#--- data-and-bss.s
51+
.globl main
52+
main:
53+
nop
54+
55+
.data
56+
.quad 1
57+
.byte 2
58+
59+
.section .data_cygwin_nocopy, "w"
60+
.align 4
61+
.quad 3
62+
.byte 4
63+
64+
.bss
65+
.zero 8192
66+
67+
.section .test, "w"
68+
.quad __data_start__
69+
.quad __data_end__
70+
.quad __bss_start__
71+
.quad __bss_end__
72+
73+
# DATANOBSS: Contents of section .data:
74+
# DATANOBSS-NEXT: 140003000 01000000 00000000 02000000 03000000
75+
# DATANOBSS-NEXT: 140003010 00000000 04
76+
# __data_start__ pointing at 0x140003000 and
77+
# __data_end__ pointing at 0x140003009.
78+
# DATANOBSS-NEXT: Contents of section .test:
79+
# DATANOBSS-NEXT: 140004000 00300040 01000000 09300040 01000000
80+
# DATANOBSS-NEXT: 140004010 18300040 01000000 18300040 01000000
81+
82+
# __bss_start__ pointing at 0x140003000 and
83+
# __bss_end__ pointing at 0x140005000.
84+
# BSSNODATA-NOT: Contents of section .data:
85+
# BSSNODATA: Contents of section .test:
86+
# BSSNODATA-NEXT: 140005000 00300040 01000000 00300040 01000000
87+
# BSSNODATA-NEXT: 140005010 00300040 01000000 00500040 01000000
88+
89+
# DATAANDBSS: Contents of section .data:
90+
# DATAANDBSS-NEXT: 140003000 01000000 00000000 02000000 03000000
91+
# DATAANDBSS-NEXT: 140003010 00000000 04000000 00000000 00000000
92+
# __data_start__ pointing at 0x140003000 and
93+
# __data_end__ pointing at 0x140003009.
94+
# __bss_start__ pointing at 0x140003018 and
95+
# __bss_end__ pointing at 0x140005018.
96+
# DATAANDBSS: 1400031f0 00000000 00000000 00000000 00000000
97+
# DATAANDBSS-NEXT: Contents of section .test:
98+
# DATAANDBSS-NEXT: 140006000 00300040 01000000 09300040 01000000
99+
# DATAANDBSS-NEXT: 140006010 18300040 01000000 18500040 01000000

0 commit comments

Comments
 (0)