Do not store R31 into the caller's link area on PPC.

This violates the ABI (that area is "reserved"), and
while it is safe if all code is generated with current
compilers, there is some very old code around that uses
that slot for something else, and breaks if it is stored
into.  Adjust testcases looking for current behavior.
I've verified that the stack frame size is right in all
testcases, whether it changed or not.  7311323.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@89811 91177308-0d34-0410-b5e6-96231b3b80d8

6 files changed, 43 insertions, 32 deletions

diff --git a/lib/Target/PowerPC/PPCFrameInfo.h b/lib/Target/PowerPC/PPCFrameInfo.h
index 65f113e6fb..4bf543a8d0 100644
--- a/lib/Target/PowerPC/PPCFrameInfo.h
+++ b/lib/Target/PowerPC/PPCFrameInfo.h
@@ -42,11 +42,12 @@ public:
   /// frame pointer.
   static unsigned getFramePointerSaveOffset(bool isPPC64, bool isDarwinABI) {
     // For the Darwin ABI:
-    // Use the TOC save slot in the PowerPC linkage area for saving the frame
-    // pointer (if needed.)  LLVM does not generate code that uses the TOC (R2
-    // is treated as a caller saved register.)
+    // We cannot use the TOC save slot (offset +20) in the PowerPC linkage area
+    // for saving the frame pointer (if needed.)  While the published ABI has
+    // not used this slot since at least MacOSX 10.2, there is older code
+    // around that does use it, and that needs to continue to work.
     if (isDarwinABI)
-      return isPPC64 ? 40 : 20;
+      return isPPC64 ? -8U : -4U;
     
     // SVR4 ABI: First slot in the general register save area.
     return -4U;
@@ -90,6 +91,17 @@ public:
   // With the SVR4 ABI, callee-saved registers have fixed offsets on the stack.
   const SpillSlot *
   getCalleeSavedSpillSlots(unsigned &NumEntries) const {
+    if (TM.getSubtarget<PPCSubtarget>().isDarwinABI()) {
+      NumEntries = 1;
+      if (TM.getSubtarget<PPCSubtarget>().isPPC64()) {
+        static const SpillSlot darwin64Offsets[] = {PPC::X31, -8};
+        return darwin64Offsets;
+      } else {
+        static const SpillSlot darwinOffsets[] = {PPC::R31, -4};
+        return darwinOffsets;
+      }
+    }
+
     // Early exit if not using the SVR4 ABI.
     if (!TM.getSubtarget<PPCSubtarget>().isSVR4ABI()) {
       NumEntries = 0;
diff --git a/lib/Target/PowerPC/PPCRegisterInfo.cpp b/lib/Target/PowerPC/PPCRegisterInfo.cpp
index 9a1a00f462..0c3c8eb649 100644
--- a/lib/Target/PowerPC/PPCRegisterInfo.cpp
+++ b/lib/Target/PowerPC/PPCRegisterInfo.cpp
@@ -1033,12 +1033,11 @@ PPCRegisterInfo::processFunctionBeforeCalleeSavedScan(MachineFunction &MF,
   //  Save R31 if necessary
   int FPSI = FI->getFramePointerSaveIndex();
   bool isPPC64 = Subtarget.isPPC64();
-  bool isSVR4ABI = Subtarget.isSVR4ABI();
   bool isDarwinABI  = Subtarget.isDarwinABI();
   MachineFrameInfo *MFI = MF.getFrameInfo();
  
   // If the frame pointer save index hasn't been defined yet.
-  if (!FPSI && needsFP(MF) && isSVR4ABI) {
+  if (!FPSI && needsFP(MF)) {
     // Find out what the fix offset of the frame pointer save area.
     int FPOffset = PPCFrameInfo::getFramePointerSaveOffset(isPPC64,
                                                            isDarwinABI);
diff --git a/test/CodeGen/PowerPC/Frames-alloca.ll b/test/CodeGen/PowerPC/Frames-alloca.ll
index 25fc626550..aed4fdbb2d 100644
--- a/test/CodeGen/PowerPC/Frames-alloca.ll
+++ b/test/CodeGen/PowerPC/Frames-alloca.ll
@@ -6,23 +6,23 @@
 ; RUN: llc < %s -march=ppc32 -mtriple=powerpc-apple-darwin8 -enable-ppc32-regscavenger | FileCheck %s -check-prefix=PPC32-RS
 ; RUN: llc < %s -march=ppc32 -mtriple=powerpc-apple-darwin8 -disable-fp-elim -enable-ppc32-regscavenger | FileCheck %s -check-prefix=PPC32-RS-NOFP
 
-; CHECK-PPC32: stw r31, 20(r1)
+; CHECK-PPC32: stw r31, -4(r1)
 ; CHECK-PPC32: lwz r1, 0(r1)
-; CHECK-PPC32: lwz r31, 20(r1)
-; CHECK-PPC32-NOFP: stw r31, 20(r1)
+; CHECK-PPC32: lwz r31, -4(r1)
+; CHECK-PPC32-NOFP: stw r31, -4(r1)
 ; CHECK-PPC32-NOFP: lwz r1, 0(r1)
-; CHECK-PPC32-NOFP: lwz r31, 20(r1)
+; CHECK-PPC32-NOFP: lwz r31, -4(r1)
 ; CHECK-PPC32-RS: stwu r1, -80(r1)
 ; CHECK-PPC32-RS-NOFP: stwu r1, -80(r1)
 
-; CHECK-PPC64: std r31, 40(r1)
-; CHECK-PPC64: stdu r1, -112(r1)
+; CHECK-PPC64: std r31, -8(r1)
+; CHECK-PPC64: stdu r1, -128(r1)
 ; CHECK-PPC64: ld r1, 0(r1)
-; CHECK-PPC64: ld r31, 40(r1)
-; CHECK-PPC64-NOFP: std r31, 40(r1)
-; CHECK-PPC64-NOFP: stdu r1, -112(r1)
+; CHECK-PPC64: ld r31, -8(r1)
+; CHECK-PPC64-NOFP: std r31, -8(r1)
+; CHECK-PPC64-NOFP: stdu r1, -128(r1)
 ; CHECK-PPC64-NOFP: ld r1, 0(r1)
-; CHECK-PPC64-NOFP: ld r31, 40(r1)
+; CHECK-PPC64-NOFP: ld r31, -8(r1)
 
 define i32* @f1(i32 %n) {
 	%tmp = alloca i32, i32 %n		; <i32*> [#uses=1]
diff --git a/test/CodeGen/PowerPC/Frames-large.ll b/test/CodeGen/PowerPC/Frames-large.ll
index fda2e4ff9c..302d3df284 100644
--- a/test/CodeGen/PowerPC/Frames-large.ll
+++ b/test/CodeGen/PowerPC/Frames-large.ll
@@ -22,13 +22,13 @@ define i32* @f1() nounwind {
 ; PPC32-NOFP: 	blr 
 
 ; PPC32-FP: _f1:
-; PPC32-FP:	stw r31, 20(r1)
+; PPC32-FP:	stw r31, -4(r1)
 ; PPC32-FP:	lis r0, -1
 ; PPC32-FP:	ori r0, r0, 32704
 ; PPC32-FP:	stwux r1, r1, r0
 ; ...
 ; PPC32-FP:	lwz r1, 0(r1)
-; PPC32-FP:	lwz r31, 20(r1)
+; PPC32-FP:	lwz r31, -4(r1)
 ; PPC32-FP:	blr 
 
 
@@ -42,11 +42,11 @@ define i32* @f1() nounwind {
 
 
 ; PPC64-FP: _f1:
-; PPC64-FP:	std r31, 40(r1)
+; PPC64-FP:	std r31, -8(r1)
 ; PPC64-FP:	lis r0, -1
-; PPC64-FP:	ori r0, r0, 32656
+; PPC64-FP:	ori r0, r0, 32640
 ; PPC64-FP:	stdux r1, r1, r0
 ; ...
 ; PPC64-FP:	ld r1, 0(r1)
-; PPC64-FP:	ld r31, 40(r1)
+; PPC64-FP:	ld r31, -8(r1)
 ; PPC64-FP:	blr 
diff --git a/test/CodeGen/PowerPC/Frames-small.ll b/test/CodeGen/PowerPC/Frames-small.ll
index 6875704cf3..404fdd0196 100644
--- a/test/CodeGen/PowerPC/Frames-small.ll
+++ b/test/CodeGen/PowerPC/Frames-small.ll
@@ -1,26 +1,26 @@
 ; RUN: llc < %s -march=ppc32 -mtriple=powerpc-apple-darwin8 -o %t1
-; RUN  not grep {stw r31, 20(r1)} %t1
+; RUN  not grep {stw r31, -4(r1)} %t1
 ; RUN: grep {stwu r1, -16448(r1)} %t1
 ; RUN: grep {addi r1, r1, 16448} %t1
 ; RUN: llc < %s -march=ppc32 | \
-; RUN: not grep {lwz r31, 20(r1)}
+; RUN: not grep {lwz r31, -4(r1)}
 ; RUN: llc < %s -march=ppc32 -mtriple=powerpc-apple-darwin8 -disable-fp-elim \
 ; RUN:   -o %t2
-; RUN: grep {stw r31, 20(r1)} %t2
+; RUN: grep {stw r31, -4(r1)} %t2
 ; RUN: grep {stwu r1, -16448(r1)} %t2
 ; RUN: grep {addi r1, r1, 16448} %t2
-; RUN: grep {lwz r31, 20(r1)} %t2
+; RUN: grep {lwz r31, -4(r1)} %t2
 ; RUN: llc < %s -march=ppc64 -mtriple=powerpc-apple-darwin8 -o %t3
-; RUN: not grep {std r31, 40(r1)} %t3
+; RUN: not grep {std r31, -8(r1)} %t3
 ; RUN: grep {stdu r1, -16496(r1)} %t3
 ; RUN: grep {addi r1, r1, 16496} %t3
-; RUN: not grep {ld r31, 40(r1)} %t3
+; RUN: not grep {ld r31, -8(r1)} %t3
 ; RUN: llc < %s -march=ppc64 -mtriple=powerpc-apple-darwin8 -disable-fp-elim \
 ; RUN:   -o %t4
-; RUN: grep {std r31, 40(r1)} %t4
-; RUN: grep {stdu r1, -16496(r1)} %t4
-; RUN: grep {addi r1, r1, 16496} %t4
-; RUN: grep {ld r31, 40(r1)} %t4
+; RUN: grep {std r31, -8(r1)} %t4
+; RUN: grep {stdu r1, -16512(r1)} %t4
+; RUN: grep {addi r1, r1, 16512} %t4
+; RUN: grep {ld r31, -8(r1)} %t4
 
 define i32* @f1() {
         %tmp = alloca i32, i32 4095             ; <i32*> [#uses=1]
diff --git a/test/CodeGen/PowerPC/ppc-prologue.ll b/test/CodeGen/PowerPC/ppc-prologue.ll
index 581d010a90..e49dcb82c6 100644
--- a/test/CodeGen/PowerPC/ppc-prologue.ll
+++ b/test/CodeGen/PowerPC/ppc-prologue.ll
@@ -2,7 +2,7 @@
 
 define i32 @_Z4funci(i32 %a) ssp {
 ; CHECK:       mflr r0
-; CHECK-NEXT:  stw r31, 20(r1)
+; CHECK-NEXT:  stw r31, -4(r1)
 ; CHECK-NEXT:  stw r0, 8(r1)
 ; CHECK-NEXT:  stwu r1, -80(r1)
 ; CHECK-NEXT: Llabel1: