[LLD][COFF] Handle emulated-TLS symbols generated by LTO #145310
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
LLD normally performs auto-import and auto-export analysis before LTO, because LTO needs to know which symbols should be preserved. However, when using emulated-TLS mode, LTO runs a LLVM pass named `LowerEmuTLS`, which transforms the TLS variables. This can result in symbols being added or removed after the initial analysis: - Symbols generated by `LowerEmuTLS` (e.g. `__emutls_v.var`) are not marked for import or export. - Original TLS variables (`var`) removed by `LowerEmuTLS` may still be referenced by the GC, leading to `undefined symbol` errors during linking. This patch introduces special handling around LTO for the emulated-TLS case: 1. Skip checking for unresolvable symbols since symbol table may be modified by LTO. 2. Skip auto-export before LTO; instead, mark symbols eligible for auto-export as "used" so they are preserved during LTO. 3. If `__emutls_v.var` is marked as used, also mark `var` as used, so it isn't prematurely discarded by LTO. 4. Re-run auto-importing after LTO, as the generated emulated-TLS symbols may need to be auto-imported. 5. Run auto-exporting after post-LTO auto-importing, to ensure newly generated symbols are exported properly. These changes are applied only when both LTO and emulated TLS are enabled.
|
@llvm/pr-subscribers-lld-coff @llvm/pr-subscribers-lld Author: Tomohiro Kashiwada (kikairoya) ChangesLLD normally performs auto-import and auto-export analysis before LTO, because LTO needs to know which symbols should be preserved. However, when using emulated-TLS mode, LTO runs a LLVM pass named
This patch introduces special handling around LTO for the emulated-TLS case:
These changes are applied only when both LTO and emulated TLS are enabled. Full diff: https://github.com/llvm/llvm-project/pull/145310.diff 3 Files Affected:
diff --git a/lld/COFF/Driver.cpp b/lld/COFF/Driver.cpp
index f3240b22a1442..bdd0886937c04 100644
--- a/lld/COFF/Driver.cpp
+++ b/lld/COFF/Driver.cpp
@@ -1384,7 +1384,8 @@ void LinkerDriver::pullArm64ECIcallHelper() {
// explicitly specified. The automatic behavior can be disabled using the
// -exclude-all-symbols option, so that lld-link behaves like link.exe rather
// than MinGW in the case that nothing is explicitly exported.
-void LinkerDriver::maybeExportMinGWSymbols(const opt::InputArgList &args) {
+void LinkerDriver::maybeExportMinGWSymbols(const opt::InputArgList &args,
+ bool preLTOCheck) {
if (!args.hasArg(OPT_export_all_symbols)) {
if (!ctx.config.dll)
return;
@@ -1415,7 +1416,7 @@ void LinkerDriver::maybeExportMinGWSymbols(const opt::InputArgList &args) {
if (!exporter.shouldExport(def))
return;
- if (!def->isGCRoot) {
+ if (!preLTOCheck && !def->isGCRoot) {
def->isGCRoot = true;
ctx.config.gcroot.push_back(def);
}
@@ -1427,7 +1428,8 @@ void LinkerDriver::maybeExportMinGWSymbols(const opt::InputArgList &args) {
if (!(c->getOutputCharacteristics() & IMAGE_SCN_MEM_EXECUTE))
e.data = true;
s->isUsedInRegularObj = true;
- symtab.exports.push_back(e);
+ if (!preLTOCheck)
+ symtab.exports.push_back(e);
});
});
}
@@ -2580,6 +2582,16 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
ctx.forEachActiveSymtab(
[&](SymbolTable &symtab) { symtab.addUndefinedGlob(pat); });
+ // combination of LTO and EmuTLS needs special handlings
+ const bool isEmuTLS = llvm::is_contained(config->mllvmOpts, "-emulated-tls");
+ const bool isLTO = [&]() {
+ bool isLTO = false;
+ ctx.forEachSymtab([&](SymbolTable &symtab) {
+ isLTO |= !symtab.bitcodeFileInstances.empty();
+ });
+ return isLTO;
+ }();
+
// Create wrapped symbols for -wrap option.
ctx.forEachSymtab([&](SymbolTable &symtab) {
addWrappedSymbols(symtab, args);
@@ -2618,11 +2630,11 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
// If we are going to do codegen for link-time optimization, check for
// unresolvable symbols first, so we don't spend time generating code that
// will fail to link anyway.
- if (!config->forceUnresolved)
- ctx.forEachSymtab([](SymbolTable &symtab) {
- if (!symtab.bitcodeFileInstances.empty())
- symtab.reportUnresolvable();
- });
+ // This check cannot be appllied if emulated TLS enabled since LTO invokes
+ // LLVM passes that modify symbol set.
+ if (isLTO && !isEmuTLS && !config->forceUnresolved)
+ ctx.forEachSymtab(
+ [&](SymbolTable &symtab) { symtab.reportUnresolvable(); });
if (errorCount())
return;
@@ -2632,7 +2644,20 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
if (config->mingw) {
// In MinGW, all symbols are automatically exported if no symbols
// are chosen to be exported.
- maybeExportMinGWSymbols(args);
+ maybeExportMinGWSymbols(args, isLTO && isEmuTLS);
+ }
+
+ // retain VAR if __emutls_v.VAR is retained
+ if (isLTO && isEmuTLS) {
+ ctx.forEachActiveSymtab([&](SymbolTable &symtab) {
+ symtab.forEachSymbol([&](Symbol *s) {
+ auto name = s->getName();
+ if (!name.starts_with("__emutls_v.") || !s->isUsedInRegularObj)
+ return;
+ if (Symbol *tls = symtab.find(name.substr(strlen("__emutls_v."))))
+ tls->isUsedInRegularObj = true;
+ });
+ });
}
// Do LTO by compiling bitcode input files to a set of native COFF files then
@@ -2641,6 +2666,15 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
ltoCompilationDone = true;
ctx.forEachSymtab([](SymbolTable &symtab) { symtab.compileBitcodeFiles(); });
+ // When emulated TLS enabled, LTO modifies symbol set. Secondly check is
+ // needed here.
+ if (isLTO && isEmuTLS && config->mingw) {
+ if (config->autoImport || config->stdcallFixup) {
+ ctx.forEachSymtab([](SymbolTable &symtab) { symtab.loadMinGWSymbols(); });
+ }
+ maybeExportMinGWSymbols(args, false);
+ }
+
if (Defined *d =
dyn_cast_or_null<Defined>(ctx.symtab.findUnderscore("_tls_used")))
config->gcroot.push_back(d);
diff --git a/lld/COFF/Driver.h b/lld/COFF/Driver.h
index 14c97a98875bf..80c4a032d0bfc 100644
--- a/lld/COFF/Driver.h
+++ b/lld/COFF/Driver.h
@@ -161,7 +161,8 @@ class LinkerDriver {
// trees into one resource tree.
void convertResources();
- void maybeExportMinGWSymbols(const llvm::opt::InputArgList &args);
+ void maybeExportMinGWSymbols(const llvm::opt::InputArgList &args,
+ bool preLTOCheck);
// We don't want to add the same file more than once.
// Files are uniquified by their filesystem and file number.
diff --git a/lld/test/COFF/lto-emutls-dll.ll b/lld/test/COFF/lto-emutls-dll.ll
new file mode 100644
index 0000000000000..3014cc0f9f811
--- /dev/null
+++ b/lld/test/COFF/lto-emutls-dll.ll
@@ -0,0 +1,133 @@
+; REQUIRES: x86
+; RUN: split-file %s %t.dir
+
+; RUN: llvm-as %t.dir/main.ll -o %t.main.bc
+; RUN: llvm-as %t.dir/other.ll -o %t.other.bc
+; RUN: llc --emulated-tls %t.dir/main.ll -o %t.main.obj --filetype=obj
+; RUN: llc --emulated-tls %t.dir/other.ll -o %t.other.obj --filetype=obj
+; RUN: llc --emulated-tls %t.dir/runtime.ll -o %t.runtime.obj --filetype=obj
+
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.obj %t.other.obj %t.runtime.obj -entry __main -o %t.exe
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.bc %t.other.bc %t.runtime.obj -entry __main -o %t.exe
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.bc %t.other.obj %t.runtime.obj -entry __main -o %t.exe
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.obj %t.other.bc %t.runtime.obj -entry __main -o %t.exe
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls -shared %t.other.bc %t.runtime.obj -o %t.dll --out-implib=%t.lib
+; RUN: llvm-readobj --coff-exports %t.dll | FileCheck %s
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.bc %t.dll %t.runtime.obj -entry __main -o %t.exe
+; RUN: llvm-readobj --coff-imports %t.exe | FileCheck %s
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.bc %t.lib %t.runtime.obj -entry __main -o %t.exe
+; RUN: llvm-readobj --coff-imports %t.exe | FileCheck %s
+
+; CHECK: _Z11set_tls_vari
+; CHECK: __emutls_v.tls_var
+
+;--- main.ll
+;; generated from:
+;;;extern int thread_local tls_var;
+;;;void set_tls_var(int v);
+;;;int main(int argc, char **argv) {
+;;; set_tls_var(3);
+;;; return tls_var == argc;
+;;;}
+target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-w64-windows-gnu"
+
+$_ZTW7tls_var = comdat any
+
+@tls_var = external dso_local thread_local global i32, align 4
+
+; Function Attrs: mustprogress noinline norecurse optnone uwtable
+define dso_local noundef i32 @__main(i32 noundef %0, ptr noundef %1) #0 {
+ %3 = alloca i32, align 4
+ %4 = alloca i32, align 4
+ %5 = alloca ptr, align 8
+ store i32 0, ptr %3, align 4
+ store i32 %0, ptr %4, align 4
+ store ptr %1, ptr %5, align 8
+ call void @_Z11set_tls_vari(i32 noundef 3)
+ %6 = call ptr @_ZTW7tls_var()
+ %7 = load i32, ptr %6, align 4
+ %8 = load i32, ptr %4, align 4
+ %9 = icmp eq i32 %7, %8
+ %10 = zext i1 %9 to i32
+ ret i32 %10
+}
+
+declare dso_local void @_Z11set_tls_vari(i32 noundef) #1
+
+; Function Attrs: noinline uwtable
+define linkonce_odr hidden noundef ptr @_ZTW7tls_var() #2 comdat {
+ %1 = icmp ne ptr @_ZTH7tls_var, null
+ br i1 %1, label %2, label %3
+
+2: ; preds = %0
+ call void @_ZTH7tls_var()
+ br label %3
+
+3: ; preds = %2, %0
+ %4 = call align 4 ptr @llvm.threadlocal.address.p0(ptr align 4 @tls_var)
+ ret ptr %4
+}
+
+declare extern_weak void @_ZTH7tls_var() #1
+
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare nonnull ptr @llvm.threadlocal.address.p0(ptr nonnull) #3
+
+!llvm.module.flags = !{!1}
+
+!1 = !{i32 1, !"ThinLTO", i32 0}
+
+;--- other.ll
+;; generated from:
+;;;int thread_local tls_var;
+;;;void set_tls_var(int v) { tls_var = v; }
+target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-w64-windows-gnu"
+
+$_ZTW7tls_var = comdat any
+
+@tls_var = dso_local thread_local global i32 0, align 4
+
+; Function Attrs: mustprogress noinline nounwind optnone uwtable
+define dso_local void @_Z11set_tls_vari(i32 noundef %0) #0 {
+ %2 = alloca i32, align 4
+ store i32 %0, ptr %2, align 4
+ %3 = load i32, ptr %2, align 4
+ %4 = call align 4 ptr @llvm.threadlocal.address.p0(ptr align 4 @tls_var)
+ store i32 %3, ptr %4, align 4
+ ret void
+}
+
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare nonnull ptr @llvm.threadlocal.address.p0(ptr nonnull) #1
+
+; Function Attrs: noinline uwtable
+define weak_odr hidden noundef ptr @_ZTW7tls_var() #2 comdat {
+ %1 = call align 4 ptr @llvm.threadlocal.address.p0(ptr align 4 @tls_var)
+ ret ptr %1
+}
+
+!llvm.module.flags = !{!1}
+
+!1 = !{i32 1, !"ThinLTO", i32 0}
+
+;--- runtime.ll
+target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-w64-windows-gnu"
+
+define dso_local hidden noundef i32 @__emutls_get_address() {
+ ret i32 0
+}
+
+define dso_local hidden noundef i32 @__emutls_register_common() {
+ ret i32 0
+}
+
+define dso_local hidden noundef i32 @_pei386_runtime_relocator() {
+ ret i32 0
+}
+
+define dso_local hidden noundef i32 @_DllMainCRTStartup() {
+ ret i32 0
+}
|
|
@llvm/pr-subscribers-platform-windows Author: Tomohiro Kashiwada (kikairoya) ChangesLLD normally performs auto-import and auto-export analysis before LTO, because LTO needs to know which symbols should be preserved. However, when using emulated-TLS mode, LTO runs a LLVM pass named
This patch introduces special handling around LTO for the emulated-TLS case:
These changes are applied only when both LTO and emulated TLS are enabled. Full diff: https://github.com/llvm/llvm-project/pull/145310.diff 3 Files Affected:
diff --git a/lld/COFF/Driver.cpp b/lld/COFF/Driver.cpp
index f3240b22a1442..bdd0886937c04 100644
--- a/lld/COFF/Driver.cpp
+++ b/lld/COFF/Driver.cpp
@@ -1384,7 +1384,8 @@ void LinkerDriver::pullArm64ECIcallHelper() {
// explicitly specified. The automatic behavior can be disabled using the
// -exclude-all-symbols option, so that lld-link behaves like link.exe rather
// than MinGW in the case that nothing is explicitly exported.
-void LinkerDriver::maybeExportMinGWSymbols(const opt::InputArgList &args) {
+void LinkerDriver::maybeExportMinGWSymbols(const opt::InputArgList &args,
+ bool preLTOCheck) {
if (!args.hasArg(OPT_export_all_symbols)) {
if (!ctx.config.dll)
return;
@@ -1415,7 +1416,7 @@ void LinkerDriver::maybeExportMinGWSymbols(const opt::InputArgList &args) {
if (!exporter.shouldExport(def))
return;
- if (!def->isGCRoot) {
+ if (!preLTOCheck && !def->isGCRoot) {
def->isGCRoot = true;
ctx.config.gcroot.push_back(def);
}
@@ -1427,7 +1428,8 @@ void LinkerDriver::maybeExportMinGWSymbols(const opt::InputArgList &args) {
if (!(c->getOutputCharacteristics() & IMAGE_SCN_MEM_EXECUTE))
e.data = true;
s->isUsedInRegularObj = true;
- symtab.exports.push_back(e);
+ if (!preLTOCheck)
+ symtab.exports.push_back(e);
});
});
}
@@ -2580,6 +2582,16 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
ctx.forEachActiveSymtab(
[&](SymbolTable &symtab) { symtab.addUndefinedGlob(pat); });
+ // combination of LTO and EmuTLS needs special handlings
+ const bool isEmuTLS = llvm::is_contained(config->mllvmOpts, "-emulated-tls");
+ const bool isLTO = [&]() {
+ bool isLTO = false;
+ ctx.forEachSymtab([&](SymbolTable &symtab) {
+ isLTO |= !symtab.bitcodeFileInstances.empty();
+ });
+ return isLTO;
+ }();
+
// Create wrapped symbols for -wrap option.
ctx.forEachSymtab([&](SymbolTable &symtab) {
addWrappedSymbols(symtab, args);
@@ -2618,11 +2630,11 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
// If we are going to do codegen for link-time optimization, check for
// unresolvable symbols first, so we don't spend time generating code that
// will fail to link anyway.
- if (!config->forceUnresolved)
- ctx.forEachSymtab([](SymbolTable &symtab) {
- if (!symtab.bitcodeFileInstances.empty())
- symtab.reportUnresolvable();
- });
+ // This check cannot be appllied if emulated TLS enabled since LTO invokes
+ // LLVM passes that modify symbol set.
+ if (isLTO && !isEmuTLS && !config->forceUnresolved)
+ ctx.forEachSymtab(
+ [&](SymbolTable &symtab) { symtab.reportUnresolvable(); });
if (errorCount())
return;
@@ -2632,7 +2644,20 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
if (config->mingw) {
// In MinGW, all symbols are automatically exported if no symbols
// are chosen to be exported.
- maybeExportMinGWSymbols(args);
+ maybeExportMinGWSymbols(args, isLTO && isEmuTLS);
+ }
+
+ // retain VAR if __emutls_v.VAR is retained
+ if (isLTO && isEmuTLS) {
+ ctx.forEachActiveSymtab([&](SymbolTable &symtab) {
+ symtab.forEachSymbol([&](Symbol *s) {
+ auto name = s->getName();
+ if (!name.starts_with("__emutls_v.") || !s->isUsedInRegularObj)
+ return;
+ if (Symbol *tls = symtab.find(name.substr(strlen("__emutls_v."))))
+ tls->isUsedInRegularObj = true;
+ });
+ });
}
// Do LTO by compiling bitcode input files to a set of native COFF files then
@@ -2641,6 +2666,15 @@ void LinkerDriver::linkerMain(ArrayRef<const char *> argsArr) {
ltoCompilationDone = true;
ctx.forEachSymtab([](SymbolTable &symtab) { symtab.compileBitcodeFiles(); });
+ // When emulated TLS enabled, LTO modifies symbol set. Secondly check is
+ // needed here.
+ if (isLTO && isEmuTLS && config->mingw) {
+ if (config->autoImport || config->stdcallFixup) {
+ ctx.forEachSymtab([](SymbolTable &symtab) { symtab.loadMinGWSymbols(); });
+ }
+ maybeExportMinGWSymbols(args, false);
+ }
+
if (Defined *d =
dyn_cast_or_null<Defined>(ctx.symtab.findUnderscore("_tls_used")))
config->gcroot.push_back(d);
diff --git a/lld/COFF/Driver.h b/lld/COFF/Driver.h
index 14c97a98875bf..80c4a032d0bfc 100644
--- a/lld/COFF/Driver.h
+++ b/lld/COFF/Driver.h
@@ -161,7 +161,8 @@ class LinkerDriver {
// trees into one resource tree.
void convertResources();
- void maybeExportMinGWSymbols(const llvm::opt::InputArgList &args);
+ void maybeExportMinGWSymbols(const llvm::opt::InputArgList &args,
+ bool preLTOCheck);
// We don't want to add the same file more than once.
// Files are uniquified by their filesystem and file number.
diff --git a/lld/test/COFF/lto-emutls-dll.ll b/lld/test/COFF/lto-emutls-dll.ll
new file mode 100644
index 0000000000000..3014cc0f9f811
--- /dev/null
+++ b/lld/test/COFF/lto-emutls-dll.ll
@@ -0,0 +1,133 @@
+; REQUIRES: x86
+; RUN: split-file %s %t.dir
+
+; RUN: llvm-as %t.dir/main.ll -o %t.main.bc
+; RUN: llvm-as %t.dir/other.ll -o %t.other.bc
+; RUN: llc --emulated-tls %t.dir/main.ll -o %t.main.obj --filetype=obj
+; RUN: llc --emulated-tls %t.dir/other.ll -o %t.other.obj --filetype=obj
+; RUN: llc --emulated-tls %t.dir/runtime.ll -o %t.runtime.obj --filetype=obj
+
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.obj %t.other.obj %t.runtime.obj -entry __main -o %t.exe
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.bc %t.other.bc %t.runtime.obj -entry __main -o %t.exe
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.bc %t.other.obj %t.runtime.obj -entry __main -o %t.exe
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.obj %t.other.bc %t.runtime.obj -entry __main -o %t.exe
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls -shared %t.other.bc %t.runtime.obj -o %t.dll --out-implib=%t.lib
+; RUN: llvm-readobj --coff-exports %t.dll | FileCheck %s
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.bc %t.dll %t.runtime.obj -entry __main -o %t.exe
+; RUN: llvm-readobj --coff-imports %t.exe | FileCheck %s
+; RUN: ld.lld -m i386pep -plugin-opt=mcpu=x86-64 -plugin-opt=-emulated-tls %t.main.bc %t.lib %t.runtime.obj -entry __main -o %t.exe
+; RUN: llvm-readobj --coff-imports %t.exe | FileCheck %s
+
+; CHECK: _Z11set_tls_vari
+; CHECK: __emutls_v.tls_var
+
+;--- main.ll
+;; generated from:
+;;;extern int thread_local tls_var;
+;;;void set_tls_var(int v);
+;;;int main(int argc, char **argv) {
+;;; set_tls_var(3);
+;;; return tls_var == argc;
+;;;}
+target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-w64-windows-gnu"
+
+$_ZTW7tls_var = comdat any
+
+@tls_var = external dso_local thread_local global i32, align 4
+
+; Function Attrs: mustprogress noinline norecurse optnone uwtable
+define dso_local noundef i32 @__main(i32 noundef %0, ptr noundef %1) #0 {
+ %3 = alloca i32, align 4
+ %4 = alloca i32, align 4
+ %5 = alloca ptr, align 8
+ store i32 0, ptr %3, align 4
+ store i32 %0, ptr %4, align 4
+ store ptr %1, ptr %5, align 8
+ call void @_Z11set_tls_vari(i32 noundef 3)
+ %6 = call ptr @_ZTW7tls_var()
+ %7 = load i32, ptr %6, align 4
+ %8 = load i32, ptr %4, align 4
+ %9 = icmp eq i32 %7, %8
+ %10 = zext i1 %9 to i32
+ ret i32 %10
+}
+
+declare dso_local void @_Z11set_tls_vari(i32 noundef) #1
+
+; Function Attrs: noinline uwtable
+define linkonce_odr hidden noundef ptr @_ZTW7tls_var() #2 comdat {
+ %1 = icmp ne ptr @_ZTH7tls_var, null
+ br i1 %1, label %2, label %3
+
+2: ; preds = %0
+ call void @_ZTH7tls_var()
+ br label %3
+
+3: ; preds = %2, %0
+ %4 = call align 4 ptr @llvm.threadlocal.address.p0(ptr align 4 @tls_var)
+ ret ptr %4
+}
+
+declare extern_weak void @_ZTH7tls_var() #1
+
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare nonnull ptr @llvm.threadlocal.address.p0(ptr nonnull) #3
+
+!llvm.module.flags = !{!1}
+
+!1 = !{i32 1, !"ThinLTO", i32 0}
+
+;--- other.ll
+;; generated from:
+;;;int thread_local tls_var;
+;;;void set_tls_var(int v) { tls_var = v; }
+target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-w64-windows-gnu"
+
+$_ZTW7tls_var = comdat any
+
+@tls_var = dso_local thread_local global i32 0, align 4
+
+; Function Attrs: mustprogress noinline nounwind optnone uwtable
+define dso_local void @_Z11set_tls_vari(i32 noundef %0) #0 {
+ %2 = alloca i32, align 4
+ store i32 %0, ptr %2, align 4
+ %3 = load i32, ptr %2, align 4
+ %4 = call align 4 ptr @llvm.threadlocal.address.p0(ptr align 4 @tls_var)
+ store i32 %3, ptr %4, align 4
+ ret void
+}
+
+; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
+declare nonnull ptr @llvm.threadlocal.address.p0(ptr nonnull) #1
+
+; Function Attrs: noinline uwtable
+define weak_odr hidden noundef ptr @_ZTW7tls_var() #2 comdat {
+ %1 = call align 4 ptr @llvm.threadlocal.address.p0(ptr align 4 @tls_var)
+ ret ptr %1
+}
+
+!llvm.module.flags = !{!1}
+
+!1 = !{i32 1, !"ThinLTO", i32 0}
+
+;--- runtime.ll
+target datalayout = "e-m:w-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-w64-windows-gnu"
+
+define dso_local hidden noundef i32 @__emutls_get_address() {
+ ret i32 0
+}
+
+define dso_local hidden noundef i32 @__emutls_register_common() {
+ ret i32 0
+}
+
+define dso_local hidden noundef i32 @_pei386_runtime_relocator() {
+ ret i32 0
+}
+
+define dso_local hidden noundef i32 @_DllMainCRTStartup() {
+ ret i32 0
+}
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
LLD normally performs auto-import and auto-export analysis before LTO, because LTO needs to know which symbols should be preserved. However, when using emulated-TLS mode, LTO runs a LLVM pass named
LowerEmuTLS, which transforms the TLS variables. This can result in symbols being added or removed after the initial analysis:LowerEmuTLS(e.g.__emutls_v.var) are not marked for import or export.var) removed byLowerEmuTLSmay still be referenced by the GC, leading toundefined symbolerrors during linking.This patch introduces special handling around LTO for the emulated-TLS case:
__emutls_v.varis marked as used, also markvaras used, so it isn't prematurely discarded by LTO.These changes are applied only when both LTO and emulated TLS are enabled.
Resolves #145195 #79715