[Clang][CodeGen] Implement code generation for __builtin_infer_alloc_token() (#156842)

Implement code generation for `__builtin_infer_alloc_token()`. The
`AllocToken` pass is now registered to run unconditionally in the
optimization pipeline.  This ensures that all instances of the
`llvm.alloc.token.id` intrinsic are lowered to constant token IDs,
regardless of whether `-fsanitize=alloc-token` is enabled. This
guarantees that the builtin always resolves to a token value, providing
a consistent and reliable mechanism for compile-time token querying.

This completes `__builtin_infer_alloc_token(<malloc-args>, ...)` to
allow compile-time querying of the token ID, where the builtin arguments
mirror those normally passed to any allocation function. The argument
expressions are unevaluated operands. For type-based token modes, the
same type inference logic is used as for untyped allocation calls.

For example the ID that is passed to (with `-fsanitize=alloc-token`):

    some_malloc(sizeof(Type), ...)

is equivalent to the token ID returned by

    __builtin_infer_alloc_token(sizeof(Type), ...)

The builtin provides a mechanism to pass or compare token IDs in code
that needs to be explicitly allocation token-aware (such as inside an
allocator, or through wrapper macros).

A more concrete demonstration of __builtin_infer_alloc_token's use is
enabling type-aware Slab allocations in the Linux kernel:

  https://lore.kernel.org/all/20250825154505.1558444-1-elver@google.com/

Notably, any kind of allocation-call rewriting is a poor fit for the
Linux kernel's kmalloc-family functions, which are macros that wrap
(multiple) layers of inline and non-inline wrapper functions. Given the
Linux kernel defines its own allocation APIs, the more explicit builtin
gives the right level of control over where the type inference happens
and the resulting token is passed.
This commit is contained in:
Marco Elver 2025-10-28 16:55:29 +01:00 committed by GitHub
parent 624d4f6754
commit 8c8f2df232
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
6 changed files with 166 additions and 22 deletions

View File

@ -49,6 +49,39 @@ change or removal. These may (experimentally) be selected with ``-Xclang
* ``increment``: This mode assigns a simple, incrementally increasing token ID
to each allocation site.
The following command-line options affect generated token IDs:
* ``-falloc-token-max=<N>``
Configures the maximum number of tokens. No max by default (tokens bounded
by ``SIZE_MAX``).
Querying Token IDs with ``__builtin_infer_alloc_token``
=======================================================
For use cases where the token ID must be known at compile time, Clang provides
a builtin function:
.. code-block:: c
size_t __builtin_infer_alloc_token(<args>, ...);
This builtin returns the token ID inferred from its argument expressions, which
mirror arguments normally passed to any allocation function. The argument
expressions are **unevaluated**, so it can be used with expressions that would
have side effects without any runtime impact.
For example, it can be used as follows:
.. code-block:: c
struct MyType { ... };
void *__partition_alloc(size_t size, size_t partition);
#define partition_alloc(...) __partition_alloc(__VA_ARGS__, __builtin_infer_alloc_token(__VA_ARGS__))
void foo(void) {
MyType *x = partition_alloc(sizeof(*x));
}
Allocation Token Instrumentation
================================
@ -70,16 +103,6 @@ example:
// Instrumented:
ptr = __alloc_token_malloc(size, <token id>);
The following command-line options affect generated token IDs:
* ``-falloc-token-max=<N>``
Configures the maximum number of tokens. No max by default (tokens bounded
by ``SIZE_MAX``).
.. code-block:: console
% clang++ -fsanitize=alloc-token -falloc-token-max=512 example.cc
Runtime Interface
-----------------

View File

@ -281,6 +281,9 @@ Non-comprehensive list of changes in this release
allocator-level heap organization strategies. A feature to instrument all
allocation functions with a token ID can be enabled via the
``-fsanitize=alloc-token`` flag.
- A builtin ``__builtin_infer_alloc_token(<args>, ...)`` is provided to allow
compile-time querying of allocation token IDs, where the builtin arguments
mirror those normally passed to an allocation function.
- Clang now rejects the invalid use of ``constexpr`` with ``auto`` and an explicit type in C. (#GH163090)

View File

@ -800,16 +800,6 @@ static void addSanitizers(const Triple &TargetTriple,
MPM.addPass(DataFlowSanitizerPass(LangOpts.NoSanitizeFiles,
PB.getVirtualFileSystemPtr()));
}
if (LangOpts.Sanitize.has(SanitizerKind::AllocToken)) {
if (Level == OptimizationLevel::O0) {
// The default pass builder only infers libcall function attrs when
// optimizing, so we insert it here because we need it for accurate
// memory allocation function detection.
MPM.addPass(InferFunctionAttrsPass());
}
MPM.addPass(AllocTokenPass(getAllocTokenOptions(LangOpts, CodeGenOpts)));
}
};
if (ClSanitizeOnOptimizerEarlyEP) {
PB.registerOptimizerEarlyEPCallback(
@ -852,6 +842,23 @@ static void addSanitizers(const Triple &TargetTriple,
}
}
static void addAllocTokenPass(const Triple &TargetTriple,
const CodeGenOptions &CodeGenOpts,
const LangOptions &LangOpts, PassBuilder &PB) {
PB.registerOptimizerLastEPCallback([&](ModulePassManager &MPM,
OptimizationLevel Level,
ThinOrFullLTOPhase) {
if (Level == OptimizationLevel::O0 &&
LangOpts.Sanitize.has(SanitizerKind::AllocToken)) {
// The default pass builder only infers libcall function attrs when
// optimizing, so we insert it here because we need it for accurate
// memory allocation function detection with -fsanitize=alloc-token.
MPM.addPass(InferFunctionAttrsPass());
}
MPM.addPass(AllocTokenPass(getAllocTokenOptions(LangOpts, CodeGenOpts)));
});
}
void EmitAssemblyHelper::RunOptimizationPipeline(
BackendAction Action, std::unique_ptr<raw_pwrite_stream> &OS,
std::unique_ptr<llvm::ToolOutputFile> &ThinLinkOS, BackendConsumer *BC) {
@ -1106,6 +1113,7 @@ void EmitAssemblyHelper::RunOptimizationPipeline(
if (!IsThinLTOPostLink) {
addSanitizers(TargetTriple, CodeGenOpts, LangOpts, PB);
addKCFIPass(TargetTriple, LangOpts, PB);
addAllocTokenPass(TargetTriple, CodeGenOpts, LangOpts, PB);
}
if (std::optional<GCOVOptions> Options =

View File

@ -4506,6 +4506,15 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
return RValue::get(AI);
}
case Builtin::BI__builtin_infer_alloc_token: {
llvm::MDNode *MDN = buildAllocToken(E);
llvm::Value *MDV = MetadataAsValue::get(getLLVMContext(), MDN);
llvm::Function *F =
CGM.getIntrinsic(llvm::Intrinsic::alloc_token_id, {IntPtrTy});
llvm::CallBase *TokenID = Builder.CreateCall(F, MDV);
return RValue::get(TokenID);
}
case Builtin::BIbzero:
case Builtin::BI__builtin_bzero: {
Address Dest = EmitPointerWithAlignment(E->getArg(0));

View File

@ -32,10 +32,12 @@
// CHECK-FULL-O0-NEXT: Running pass: AlwaysInlinerPass
// CHECK-FULL-O0-NEXT: Running analysis: ProfileSummaryAnalysis
// CHECK-FULL-O0-NEXT: Running pass: CoroConditionalWrapper
// CHECK-FULL-O0-NEXT: Running pass: AllocTokenPass
// CHECK-FULL-O0-NEXT: Running analysis: OptimizationRemarkEmitterAnalysis
// CHECK-FULL-O0-NEXT: Running analysis: TargetLibraryAnalysis
// CHECK-FULL-O0-NEXT: Running pass: CanonicalizeAliasesPass
// CHECK-FULL-O0-NEXT: Running pass: NameAnonGlobalPass
// CHECK-FULL-O0-NEXT: Running pass: AnnotationRemarksPass
// CHECK-FULL-O0-NEXT: Running analysis: TargetLibraryAnalysis
// CHECK-FULL-O0-NEXT: Running pass: VerifierPass
// CHECK-FULL-O0-NEXT: Running pass: BitcodeWriterPass
@ -46,10 +48,12 @@
// CHECK-THIN-O0-NEXT: Running pass: AlwaysInlinerPass
// CHECK-THIN-O0-NEXT: Running analysis: ProfileSummaryAnalysis
// CHECK-THIN-O0-NEXT: Running pass: CoroConditionalWrapper
// CHECK-THIN-O0-NEXT: Running pass: AllocTokenPass
// CHECK-THIN-O0-NEXT: Running analysis: OptimizationRemarkEmitterAnalysis
// CHECK-THIN-O0-NEXT: Running analysis: TargetLibraryAnalysis
// CHECK-THIN-O0-NEXT: Running pass: CanonicalizeAliasesPass
// CHECK-THIN-O0-NEXT: Running pass: NameAnonGlobalPass
// CHECK-THIN-O0-NEXT: Running pass: AnnotationRemarksPass
// CHECK-THIN-O0-NEXT: Running analysis: TargetLibraryAnalysis
// CHECK-THIN-O0-NEXT: Running pass: VerifierPass
// CHECK-THIN-O0-NEXT: Running pass: ThinLTOBitcodeWriterPass

View File

@ -0,0 +1,97 @@
// To test IR generation of the builtin without evaluating the LLVM intrinsic,
// we set the mode to a stateful mode, which prohibits constant evaluation.
// RUN: %clang_cc1 -triple x86_64-linux-gnu -Werror -std=c++20 -emit-llvm -falloc-token-mode=random -disable-llvm-passes %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-CODEGEN
// RUN: %clang_cc1 -triple x86_64-linux-gnu -Werror -std=c++20 -emit-llvm -falloc-token-max=2 %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-LOWER
extern "C" void *my_malloc(unsigned long, unsigned long);
struct NoPtr {
int x;
long y;
};
struct WithPtr {
int a;
char *buf;
};
int unevaluated_fn();
// CHECK-LABEL: @_Z16test_builtin_intv(
// CHECK-CODEGEN: call i64 @llvm.alloc.token.id.i64(metadata ![[META_INT:[0-9]+]])
// CHECK-LOWER: ret i64 0
unsigned long test_builtin_int() {
return __builtin_infer_alloc_token(sizeof(1));
}
// CHECK-LABEL: @_Z16test_builtin_ptrv(
// CHECK-CODEGEN: call i64 @llvm.alloc.token.id.i64(metadata ![[META_PTR:[0-9]+]])
// CHECK-LOWER: ret i64 1
unsigned long test_builtin_ptr() {
return __builtin_infer_alloc_token(sizeof(int *));
}
// CHECK-LABEL: @_Z25test_builtin_struct_noptrv(
// CHECK-CODEGEN: call i64 @llvm.alloc.token.id.i64(metadata ![[META_NOPTR:[0-9]+]])
// CHECK-LOWER: ret i64 0
unsigned long test_builtin_struct_noptr() {
return __builtin_infer_alloc_token(sizeof(NoPtr));
}
// CHECK-LABEL: @_Z25test_builtin_struct_w_ptrv(
// CHECK-CODEGEN: call i64 @llvm.alloc.token.id.i64(metadata ![[META_WITHPTR:[0-9]+]])
// CHECK-LOWER: ret i64 1
unsigned long test_builtin_struct_w_ptr() {
return __builtin_infer_alloc_token(sizeof(WithPtr), 123);
}
// CHECK-LABEL: @_Z24test_builtin_unevaluatedv(
// CHECK-NOT: call{{.*}}unevaluated_fn
// CHECK-CODEGEN: call i64 @llvm.alloc.token.id.i64(metadata ![[META_INT:[0-9]+]])
// CHECK-LOWER: ret i64 0
unsigned long test_builtin_unevaluated() {
return __builtin_infer_alloc_token(sizeof(int) * unevaluated_fn());
}
// CHECK-LABEL: @_Z36test_builtin_unsequenced_unevaluatedi(
// CHECK: add nsw
// CHECK-NOT: add nsw
// CHECK-CODEGEN: %[[REG:[0-9]+]] = call i64 @llvm.alloc.token.id.i64(metadata ![[META_UNKNOWN:[0-9]+]])
// CHECK-CODEGEN: call{{.*}}@my_malloc({{.*}}, i64 noundef %[[REG]])
// CHECK-LOWER: call{{.*}}@my_malloc({{.*}}, i64 noundef 0)
void test_builtin_unsequenced_unevaluated(int x) {
my_malloc(++x, __builtin_infer_alloc_token(++x));
}
// CHECK-LABEL: @_Z20test_builtin_unknownv(
// CHECK-CODEGEN: call i64 @llvm.alloc.token.id.i64(metadata ![[META_UNKNOWN:[0-9]+]])
// CHECK-LOWER: ret i64 0
unsigned long test_builtin_unknown() {
return __builtin_infer_alloc_token(4096);
}
// Test template instantiation.
template <typename T>
constexpr unsigned long get_token() {
return __builtin_infer_alloc_token(sizeof(T));
}
// CHECK-LABEL: @_Z13get_token_intv()
// CHECK-CODEGEN: call i64 @llvm.alloc.token.id.i64(metadata ![[META_INT]])
// CHECK-LOWER: ret i64 0
unsigned long get_token_int() {
return get_token<int>();
}
// CHECK-LABEL: @_Z13get_token_ptrv()
// CHECK-CODEGEN: call i64 @llvm.alloc.token.id.i64(metadata ![[META_PTR]])
// CHECK-LOWER: ret i64 1
unsigned long get_token_ptr() {
return get_token<int *>();
}
// CHECK-CODEGEN: ![[META_INT]] = !{!"int", i1 false}
// CHECK-CODEGEN: ![[META_PTR]] = !{!"int *", i1 true}
// CHECK-CODEGEN: ![[META_NOPTR]] = !{!"NoPtr", i1 false}
// CHECK-CODEGEN: ![[META_WITHPTR]] = !{!"WithPtr", i1 true}
// CHECK-CODEGEN: ![[META_UNKNOWN]] = !{}