Add support for reverse PInvoke callouts to the JIT (#4952)

3 files changed, 90 insertions, 38 deletions

diff --git a/src/jit/compiler.h b/src/jit/compiler.h
index 88c5965b49..50c90b669c 100644
--- a/src/jit/compiler.h
+++ b/src/jit/compiler.h
@@ -2220,6 +2220,7 @@ public :
 
 #if INLINE_NDIRECT
     unsigned            lvaInlinedPInvokeFrameVar;  // variable representing the InlinedCallFrame
+    unsigned            lvaReversePInvokeFrameVar;  // variable representing the reverse PInvoke frame
 #if FEATURE_FIXED_OUT_ARGS
     unsigned            lvaPInvokeFrameRegSaveVar;  // variable representing the RegSave for PInvoke inlining.
 #endif
@@ -3413,6 +3414,8 @@ public :
 
 #endif // !_TARGET_X86_
 
+    void                fgAddReversePInvokeEnterExit();
+
     bool                fgMoreThanOneReturnBlock();
 
     // The number of separate return points in the method.
@@ -7462,6 +7465,17 @@ public :
 #endif
         }
 
+        // true if we should use insert the REVERSE_PINVOKE_{ENTER,EXIT} helpers in the method
+        // prolog/epilog
+        inline bool         IsReversePInvoke()
+        {
+#if COR_JIT_EE_VERSION > 460
+            return (jitFlags->corJitFlags2 & CORJIT_FLG2_REVERSE_PINVOKE) != 0;
+#else
+            return false;
+#endif
+        }
+
         // true if we must generate compatible code with Jit64 quirks
         inline bool         IsJit64Compat()
         {
diff --git a/src/jit/flowgraph.cpp b/src/jit/flowgraph.cpp
index 348ba66df7..44e0274d12 100644
--- a/src/jit/flowgraph.cpp
+++ b/src/jit/flowgraph.cpp
@@ -7497,8 +7497,7 @@ GenTreePtr Compiler::fgGetCritSectOfStaticMethod()
 
 void                Compiler::fgAddSyncMethodEnterExit()
 {
-    if ((info.compFlags & CORINFO_FLG_SYNCH) == 0)
-        return;
+    assert((info.compFlags & CORINFO_FLG_SYNCH) != 0);
 
     // We need to do this transformation before funclets are created.
     assert(!fgFuncletsCreated);
@@ -7808,6 +7807,69 @@ void                Compiler::fgConvertSyncReturnToLeave(BasicBlock* block)
 
 #endif // !_TARGET_X86_
 
+//------------------------------------------------------------------------
+// fgAddReversePInvokeEnterExit: Add enter/exit calls for reverse PInvoke methods
+//
+// Arguments:
+//      None.
+//
+// Return Value:
+//      None.
+
+void Compiler::fgAddReversePInvokeEnterExit()
+{
+    assert(opts.IsReversePInvoke());
+
+    lvaReversePInvokeFrameVar = lvaGrabTempWithImplicitUse(false DEBUGARG("Reverse Pinvoke FrameVar"));
+
+    LclVarDsc* varDsc = &lvaTable[lvaReversePInvokeFrameVar];
+    varDsc->lvType = TYP_BLK;
+    varDsc->lvExactSize = eeGetEEInfo()->sizeOfReversePInvokeFrame;
+
+    GenTreePtr      tree;
+
+    // Add enter pinvoke exit callout at the start of prolog
+
+    tree = gtNewOperNode(GT_ADDR, TYP_I_IMPL, gtNewLclvNode(lvaReversePInvokeFrameVar, TYP_BLK));
+
+    tree = gtNewHelperCallNode(CORINFO_HELP_JIT_REVERSE_PINVOKE_ENTER,
+        TYP_VOID, 0,
+        gtNewArgList(tree));
+
+    fgEnsureFirstBBisScratch();
+
+    fgInsertStmtAtBeg(fgFirstBB, tree);
+
+#ifdef DEBUG
+    if (verbose)
+    {
+        printf("\nReverse PInvoke method - Add reverse pinvoke enter in first basic block [%08p]\n", dspPtr(fgFirstBB));
+        gtDispTree(tree);
+        printf("\n");
+    }
+#endif
+
+    // Add reverse pinvoke exit callout at the end of epilog
+
+    tree = gtNewOperNode(GT_ADDR, TYP_I_IMPL, gtNewLclvNode(lvaReversePInvokeFrameVar, TYP_BLK));
+
+    tree = gtNewHelperCallNode(CORINFO_HELP_JIT_REVERSE_PINVOKE_EXIT,
+        TYP_VOID, 0,
+        gtNewArgList(tree));
+
+    assert(genReturnBB != nullptr);
+
+    fgInsertStmtAtEnd(genReturnBB, tree);
+
+#ifdef DEBUG
+    if (verbose)
+    {
+        printf("\nReverse PInvoke method - Add reverse pinvoke exit in return basic block [%08p]\n", dspPtr(genReturnBB));
+        gtDispTree(tree);
+        printf("\n");
+    }
+#endif
+}
 
 /*****************************************************************************
  *
@@ -7913,12 +7975,6 @@ void                Compiler::fgAddInternal()
 
     /* Assume we will generate a single shared return sequence */
 
-    // This is the node for the oneReturn statement.
-    // It could be as simple as a CallNode if we only have
-    // only one callout. It will be a comma tree of CallNodes
-    // if we have multiple callouts.
-    //
-    GenTreePtr  oneReturnStmtNode = NULL;
     ULONG       returnWeight      = 0;
     bool        oneReturn;
     bool        allProfWeight;
@@ -7933,6 +7989,7 @@ void                Compiler::fgAddInternal()
 #if INLINE_NDIRECT
          (info.compCallUnmanaged != 0) ||
 #endif
+         opts.IsReversePInvoke() ||
          ((info.compFlags & CORINFO_FLG_SYNCH) != 0))
     {
         // We will generate only one return block
@@ -8010,7 +8067,10 @@ void                Compiler::fgAddInternal()
     // BBJ_RETURN block gets placed at the top-level, not within an EH region. (Otherwise,
     // we'd have to be really careful when creating the synchronized method try/finally
     // not to include the BBJ_RETURN block.)
-    fgAddSyncMethodEnterExit();
+    if ((info.compFlags & CORINFO_FLG_SYNCH) != 0)
+    {
+        fgAddSyncMethodEnterExit();
+    }
 #endif // !_TARGET_X86_
 
     if  (oneReturn)
@@ -8300,34 +8360,13 @@ void                Compiler::fgAddInternal()
                                        gtNewArgList(tree));
         }
 
-        /* Add the exitCrit expression to the oneReturnStmtNode */
-        if (oneReturnStmtNode != NULL)
-        {
-            //
-            // We add the newly created "tree" to op1, so we can evaluate the last added
-            // expression first.
-            //
-            oneReturnStmtNode = gtNewOperNode(GT_COMMA,
-                                              TYP_VOID,
-                                              tree,
-                                              oneReturnStmtNode);
-
-        }
-        else
-        {
-            oneReturnStmtNode = tree;
-        }
-
+        fgInsertStmtAtEnd(genReturnBB, tree);
 
 #ifdef DEBUG
         if (verbose)
         {
             printf("\nSynchronized method - Add exit expression ");
             printTreeID(tree);
-            printf(" to oneReturnStmtNode ");
-            printTreeID(oneReturnStmtNode);
-            printf("\nCurrent oneReturnStmtNode is\n");
-            gtDispTree(oneReturnStmtNode);
             printf("\n");
         }
 #endif
@@ -8371,6 +8410,11 @@ void                Compiler::fgAddInternal()
 
     }
 
+    if (opts.IsReversePInvoke())
+    {
+        fgAddReversePInvokeEnterExit();
+    }
+
     //
     //  Add 'return' expression to the return block if we made it as "oneReturn" before.
     //
@@ -8382,13 +8426,6 @@ void                Compiler::fgAddInternal()
         // Make the 'return' expression.
         //
 
-        // spill any value that is currently in the oneReturnStmtNode
-        if (oneReturnStmtNode != nullptr)
-        {
-            fgInsertStmtAtEnd(genReturnBB, oneReturnStmtNode);
-            oneReturnStmtNode = nullptr;
-        }
-
         //make sure to reload the return value as part of the return (it is saved by the "real return").
         if (genReturnLocal != BAD_VAR_NUM)
         {
diff --git a/src/jit/lclvars.cpp b/src/jit/lclvars.cpp
index 569b1ee370..bd37512f71 100644
--- a/src/jit/lclvars.cpp
+++ b/src/jit/lclvars.cpp
@@ -46,6 +46,7 @@ void                Compiler::lvaInit()
     lvaShadowSPslotsVar = BAD_VAR_NUM;
 #endif // !FEATURE_EH_FUNCLETS
     lvaInlinedPInvokeFrameVar = BAD_VAR_NUM;
+    lvaReversePInvokeFrameVar = BAD_VAR_NUM;
 #if FEATURE_FIXED_OUT_ARGS
 #if INLINE_NDIRECT
     lvaPInvokeFrameRegSaveVar = BAD_VAR_NUM;