[PATCH] D150184: [ORC] Fix race-condition in RTDyldObjectLinkingLayer::onObjEmit.

[Lang Hames via Phabricator via llvm-commits]

lhames added a comment.

This is a really nice catch. Thank you for finding this!

I suspect the `ObjectLinkingLayer` has a similar issue. I'll look into that.



================
Comment at: llvm/lib/ExecutionEngine/Orc/RTDyldObjectLinkingLayer.cpp:366-391
+  // Add to MemMgrs before potentially notifying other threads, that the object
+  // was emitted. Otherwise these other threads could already execute the code
+  // and remove the object, leading to defunct resource trackers.
+  if (auto Err = R.withResourceKeyDo(
+          [&](ResourceKey K) { MemMgrs[K].push_back(std::move(MemMgr)); })) {
+    getExecutionSession().reportError(std::move(Err));
+    R.failMaterialization();
----------------
This call to `notifyObjectLoaded` will reference the moved-from `MemMgr` argument if there are any listeners attached.

The address of the MemMgr should be recorded in a local variable for use in the `notifyObjectLoaded` call. It would also be good if `MemMgr` were removed from the `MemMgrs` vector if the call to `notifyEmitted` fails.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D150184/new/

https://reviews.llvm.org/D150184