[compiler-rt][memprof] adding free_sized/free_aligned_sized intercept…

[devnexen]

…ions.

[llvmbot]

@llvm/pr-subscribers-pgo

Author: David CARLIER (devnexen)

Changes

…ions.

---

Full diff: https://github.com/llvm/llvm-project/pull/154011.diff

1 Files Affected:

• (modified) compiler-rt/lib/memprof/memprof_malloc_linux.cpp (+18)

diff --git a/compiler-rt/lib/memprof/memprof_malloc_linux.cpp b/compiler-rt/lib/memprof/memprof_malloc_linux.cpp
index 2a028c7d0b489..68fe65475889a 100644
--- a/compiler-rt/lib/memprof/memprof_malloc_linux.cpp
+++ b/compiler-rt/lib/memprof/memprof_malloc_linux.cpp
@@ -50,6 +50,24 @@ INTERCEPTOR(void, cfree, void *ptr) {
 }
 #endif // SANITIZER_INTERCEPT_CFREE
 
+#if SANITIZER_INTERCEPT_FREE_SIZED
+INTERCEPTOR(void, free_sized, void *ptr, uptr size) {
+  if (DlsymAlloc::PointerIsMine(ptr))
+    return DlsymAlloc::Free(ptr);
+  GET_STACK_TRACE_FREE;
+  memprof_delete(ptr, size, 0, &stack, FROM_MALLOC);
+}
+#endif // SANITIZER_INTERCEPT_FREE_SIZED
+
+#if SANITIZER_INTERCEPT_FREE_ALIGNED_SIZED
+INTERCEPTOR(void, free_aligned_sized, void *ptr, uptr alignment, uptr size) {
+  if (DlsymAlloc::PointerIsMine(ptr))
+    return DlsymAlloc::Free(ptr);
+  GET_STACK_TRACE_FREE;
+  memprof_delete(ptr, size, alignment, &stack, FROM_MALLOC);
+}
+#endif // SANITIZER_INTERCEPT_FREE_ALIGNED_SIZED
+
 INTERCEPTOR(void *, malloc, uptr size) {
   if (DlsymAlloc::Use())
     return DlsymAlloc::Allocate(size);

[snehasish]

Thanks for the contribution. Is it possible to add a test which exercises these new interceptors?

[devnexen]

It s going to be tough, those are C23 standard however, in practice, libc in general haven't really implemented it (see here for a glibc patch in the work).

[snehasish]

To test the interceptors we don't actually need them to be implemented. Here is an example where I patched in your changes and modified an existing test to demonstrate that the interceptors worked. The tests pass on Linux ELF with static and dyanmic linked compiler runtime.

#154228

Can you add a new test for these based on this approach?

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[snehasish]

I think the test is passing because it intercepts free not the free_sized variants you are interested in. You can verify by adding some Printf logging to the new interceptors. For the test we can simply skip defining the body of the free_sized variants and the memprof runtime will provide the right (interceptor) variant.

[snehasish]

Suggested change

	void free_sized(void *ptr, [[maybe_unused]] size_t) { free(ptr); }
	void free_sized(void *ptr, size_t);

[snehasish]

Suggested change

	void free_aligned_sized(void *ptr, [[maybe_unused]] size_t alignment,
	[[maybe_unused]] size_t size) {
	free(ptr);
	}
	void free_aligned_sized(void *ptr, size_t alignment, size_t size);

[snehasish]

lgtm

[snehasish]

typo: alignment