commit f60ed525ce73f9b66760017ecf0505f29fd22f16
author Johannes Doerfert <johannes@jdoerfert.de> Sun Feb 14 12:25:56 2021 -0600
committer Copybara-Service <copybara-worker@google.com> Fri Feb 19 04:02:49 2021 -0800
tree 683d5b52b867e2dc0b538059223061e619a660f2
parent c1808464b75435da9dcbb703e2fcf9d20cbb36ec

[OpenMP][FIX] Avoid use of stack allocations in asynchronous calls

As reported by Guilherme Valarini [0], we used to pass stack allocations
to calls that can nowadays be asynchronous. This is arguably a problem
and it will inevitably result in UB. To remedy the situation we
allocate the locations as part of the AsyncInfoTy object. The lifetime
of that object matches what we need for now. If the synchronization is
not tied to the AsyncInfoTy object anymore we might need to have a
different buffer construct in global space.

This should be back-ported to LLVM 12 but needs slight modifications as
it is based on refactoring patches we do not need to backport.

[0] https://lists.llvm.org/pipermail/openmp-dev/2021-February/003867.html

Reviewed By: JonChesterfield

Differential Revision: https://reviews.llvm.org/D96667

GitOrigin-RevId: 2518cc65d2a7564bb49d7b806f23dd47212dcefd

libomptarget/include/omptarget.h

diff --git a/libomptarget/include/omptarget.h b/libomptarget/include/omptarget.h
index ffdf965..76180fc 100644
--- a/libomptarget/include/omptarget.h
+++ b/libomptarget/include/omptarget.h
@@ -14,6 +14,7 @@
 #ifndef _OMPTARGET_H_
 #define _OMPTARGET_H_
 
+#include <deque>
 #include <stddef.h>
 #include <stdint.h>
 
@@ -136,6 +137,10 @@
 /// associated with a libomptarget layer device. RAII semantics to avoid
 /// mistakes.
 class AsyncInfoTy {
+  /// Locations we used in (potentially) asynchronous calls which should live
+  /// as long as this AsyncInfoTy object.
+  std::deque<void *> BufferLocations;
+
   __tgt_async_info AsyncInfo;
   DeviceTy &Device;
 
@@ -151,6 +156,10 @@
   ///
   /// \returns OFFLOAD_FAIL or OFFLOAD_SUCCESS appropriately.
   int synchronize();
+
+  /// Return a void* reference with a lifetime that is at least as long as this
+  /// AsyncInfoTy object. The location can be used as intermediate buffer.
+  void *&getVoidPtrLocation();
 };
 
 /// This struct is a record of non-contiguous information

libomptarget/src/omptarget.cpp

diff --git a/libomptarget/src/omptarget.cpp b/libomptarget/src/omptarget.cpp
index 51dcfec..ddefdd8 100644
--- a/libomptarget/src/omptarget.cpp
+++ b/libomptarget/src/omptarget.cpp
@@ -31,6 +31,11 @@
   return Result;
 }
 
+void *&AsyncInfoTy::getVoidPtrLocation() {
+  BufferLocations.push_back(nullptr);
+  return BufferLocations.back();
+}
+
 /* All begin addresses for partially mapped structs must be 8-aligned in order
  * to ensure proper alignment of members. E.g.
  *
@@ -432,7 +437,8 @@
       DP("Update pointer (" DPxMOD ") -> [" DPxMOD "]\n",
          DPxPTR(PointerTgtPtrBegin), DPxPTR(TgtPtrBegin));
       uint64_t Delta = (uint64_t)HstPtrBegin - (uint64_t)HstPtrBase;
-      void *TgtPtrBase = (void *)((uint64_t)TgtPtrBegin - Delta);
+      void *&TgtPtrBase = AsyncInfo.getVoidPtrLocation();
+      TgtPtrBase = (void *)((uint64_t)TgtPtrBegin - Delta);
       int rt = Device.submitData(PointerTgtPtrBegin, &TgtPtrBase,
                                  sizeof(void *), AsyncInfo);
       if (rt != OFFLOAD_SUCCESS) {
@@ -1123,8 +1129,9 @@
         DP("Parent lambda base " DPxMOD "\n", DPxPTR(TgtPtrBase));
         uint64_t Delta = (uint64_t)HstPtrBegin - (uint64_t)HstPtrBase;
         void *TgtPtrBegin = (void *)((uintptr_t)TgtPtrBase + Delta);
-        void *PointerTgtPtrBegin = Device.getTgtPtrBegin(
-            HstPtrVal, ArgSizes[I], IsLast, false, IsHostPtr);
+        void *&PointerTgtPtrBegin = AsyncInfo.getVoidPtrLocation();
+        PointerTgtPtrBegin = Device.getTgtPtrBegin(HstPtrVal, ArgSizes[I],
+                                                   IsLast, false, IsHostPtr);
         if (!PointerTgtPtrBegin) {
           DP("No lambda captured variable mapped (" DPxMOD ") - ignored\n",
              DPxPTR(HstPtrVal));