[libc] Implement 'atexit' on the GPU correctly

Summary:
This function was never marked at supported because it was fundamentally
broken when called with multiple threads. The patch in
#83026 introduces a lock-free
stack that can be used to correctly handle enqueuing callbacks from
multiple threads. Although the previous interface tried to provide a
consistent API, this was not feasible with the needs for a lock-free
stack so I have elected to just use ifdefs. The size is fixed to
whatever we use for testing, which currently amounts to about 8KiB
dedicated for this thing, which isn't enough to be concenred about.

Depends on #83026

Author: jhuber6

libc/docs/gpu/support.rst

@@ -102,6 +102,7 @@ atol           |check|
 atoll          |check|
 exit           |check|    |check|
 abort          |check|    |check|
+atexit         |check|
 labs           |check|
 llabs          |check|
 div            |check|

libc/src/stdlib/CMakeLists.txt

@@ -363,20 +363,22 @@ add_entrypoint_object(
     libc.src.__support.OSUtil.osutil
 )
 
-add_entrypoint_object(
-  atexit
-  SRCS
-    atexit.cpp
-  HDRS
-    atexit.h
-  CXX_STANDARD
-    20 # For constinit of the atexit callback list.
-  DEPENDS
-    libc.src.__support.fixedvector
-    libc.src.__support.blockstore
-    libc.src.__support.threads.mutex
-    libc.src.__support.CPP.new
-)
+if(NOT LIBC_TARGET_OS_IS_GPU)
+  add_entrypoint_object(
+    atexit
+    SRCS
+      atexit.cpp
+    HDRS
+      atexit.h
+    CXX_STANDARD
+      20 # For constinit of the atexit callback list.
+    DEPENDS
+      libc.src.__support.fixedvector
+      libc.src.__support.blockstore
+      libc.src.__support.threads.mutex
+      libc.src.__support.CPP.new
+  )
+endif()
 
 add_entrypoint_object(
   exit
@@ -398,6 +400,13 @@ add_entrypoint_object(
 )
 
 if(LIBC_TARGET_OS_IS_GPU)
+  add_entrypoint_object(
+    atexit
+    ALIAS
+    DEPENDS
+      .${LIBC_TARGET_OS}.atexit
+  )
+
   add_entrypoint_object(
     malloc
     ALIAS

libc/src/stdlib/atexit.cpp

@@ -28,14 +28,7 @@ struct AtExitUnit {
   constexpr AtExitUnit(AtExitCallback *c, void *p) : callback(c), payload(p) {}
 };
 
-#if defined(LIBC_TARGET_ARCH_IS_GPU)
-// The GPU build cannot handle the potentially recursive definitions required by
-// the BlockStore class. Additionally, the liklihood that someone exceeds this
-// while executing on the GPU is extremely small.
-// FIXME: It is not generally safe to use 'atexit' on the GPU because the
-//        mutexes simply passthrough. We will need a lock free stack.
-using ExitCallbackList = FixedVector<AtExitUnit, 64>;
-#elif defined(LIBC_COPT_PUBLIC_PACKAGING)
+#if defined(LIBC_COPT_PUBLIC_PACKAGING)
 using ExitCallbackList = cpp::ReverseOrderBlockStore<AtExitUnit, 32>;
 #else
 // BlockStore uses dynamic memory allocation. To avoid dynamic memory

libc/src/stdlib/gpu/CMakeLists.txt

@@ -30,3 +30,16 @@ add_entrypoint_object(
     libc.include.stdlib
     libc.src.__support.RPC.rpc_client
 )
+
+add_entrypoint_object(
+  atexit
+  SRCS
+    atexit.cpp
+  HDRS
+    ../atexit.h
+  CXX_STANDARD
+    20 # For constinit of the atexit callback list.
+  DEPENDS
+    libc.include.stdlib
+    libc.src.__support.fixedstack
+)

libc/src/stdlib/gpu/atexit.cpp

@@ -0,0 +1,63 @@
+//===-- GPU implementation of atexit --------------------------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "src/stdlib/atexit.h"
+#include "src/__support/common.h"
+#include "src/__support/fixedstack.h"
+
+namespace LIBC_NAMESPACE {
+
+namespace {
+
+using AtExitCallback = void(void *);
+using StdCAtExitCallback = void(void);
+
+struct AtExitUnit {
+  AtExitCallback *callback = nullptr;
+  void *payload = nullptr;
+  constexpr AtExitUnit() = default;
+  constexpr AtExitUnit(AtExitCallback *c, void *p) : callback(c), payload(p) {}
+};
+
+// The GPU interface cannot use the standard implementation because it does not
+// support the Mutex type. Instead we use a lock free stack with a sufficiently
+// large size.
+constinit FixedStack<AtExitUnit, CALLBACK_LIST_SIZE_FOR_TESTS> exit_callbacks;
+
+void stdc_at_exit_func(void *payload) {
+  reinterpret_cast<StdCAtExitCallback *>(payload)();
+}
+
+} // namespace
+
+namespace internal {
+
+void call_exit_callbacks() {
+  AtExitUnit unit;
+  while (exit_callbacks.pop(unit))
+    unit.callback(unit.payload);
+}
+
+} // namespace internal
+
+static int add_atexit_unit(const AtExitUnit &unit) {
+  if (!exit_callbacks.push(unit))
+    return -1;
+  return 0;
+}
+
+extern "C" int __cxa_atexit(AtExitCallback *callback, void *payload, void *) {
+  return add_atexit_unit({callback, payload});
+}
+
+LLVM_LIBC_FUNCTION(int, atexit, (StdCAtExitCallback * callback)) {
+  return add_atexit_unit(
+      {&stdc_at_exit_func, reinterpret_cast<void *>(callback)});
+}
+
+} // namespace LIBC_NAMESPACE

libc/startup/gpu/amdgpu/CMakeLists.txt

@@ -5,6 +5,7 @@ add_startup_object(
   DEPENDS
     libc.src.__support.RPC.rpc_client
     libc.src.__support.GPU.utils
+    libc.src.__support.CPP.new
     libc.src.stdlib.exit
     libc.src.stdlib.atexit
   COMPILE_OPTIONS

libc/startup/gpu/nvptx/CMakeLists.txt

@@ -5,6 +5,7 @@ add_startup_object(
   DEPENDS
     libc.src.__support.RPC.rpc_client
     libc.src.__support.GPU.utils
+    libc.src.__support.CPP.new
     libc.src.stdlib.exit
     libc.src.stdlib.atexit
   COMPILE_OPTIONS