SWDEV-440746 - Fix the hostcall buffer creation

Avoid a deadlock on the host call buffer creation. Since the buffer will be
allocated in the queue thread, then use direct device memory allocation
skipping the global context lock.

Change-Id: I09b55ee03bb42ab5d320c152b52a8c842c5fdcc1


[ROCm/clr commit: 62559a6e5a]

projects/clr/rocclr/device/devhostcall.cpp

@@ -284,7 +284,10 @@ static struct Init {
   ~Init() {
     if (state == State::kInit) {
       state = State::kDestroy;
-      while (state == State::kDestroy) {}
+      // @note: Under Linux thread destruction can be delayed and
+      // ROCR may crash in a wait for event occasionally. Hence, runtime needs
+      // an early exit. The logic isn't required for Windows.
+      while (IS_LINUX && (state == State::kDestroy)) {}
     }
   }
 } kHostThreadActive;