[stack protector] Improved data layout rules, part 4

[jmagee]

Summary

Hi,

This is the third part in a series of patches that will update the stack layout
rules for SSP, specifically to support the ssp-strong rules.

The breakdown of the series is:

Patch 1 [DONE]: Split the StackProtector pass from a single cpp file,
StackProtector.cpp, into StackProtector.h and StackProtector.cpp.

• Committed in r191570.
• Review at: http://llvm-reviews.chandlerc.com/D1521

Patch 2 [DONE]: Update the StackProtector pass to do the datalayout analysis.

• Committed in r193653.
• Review at: http://llvm-reviews.chandlerc.com/D1789

Patch 3 [DONE]: Use the new layout analysis to replicate the old layout behaviour.

• Committed in r197653
• Review at: http://llvm-reviews.chandlerc.com/D2158

Patch 4 [This patch]: Implements the stricter layout rules for sspstrong and sspreq.

• This patch changes the PrologueEpilogInserter and LocalStackSlotAllocation passes to follow the extended stack layout rules for sspstrong and sspreq.
• Included are the documentation updates and final tests.

Thanks,
Josh

[jmagee]

Ping.

Note, in my original summary I said this was the "third part" of the series, but it is really the fourth and final part of my SSP stack layout patch series.

• Josh

[jmagee]

Ping. (On a Friday!)

[void]

See comments inline.

docs/LangRef.rst

1064 ↗

(On Diff #6440)

Please use a monospaced font for '>= ssp-buffer-size'. here and elsewhere.

1069 ↗

(On Diff #6440)

If these are order of how it's determined where an array goes, please number the list.

lib/CodeGen/LocalStackSlotAllocation.cpp

198 ↗

(On Diff #6440)

Please use one variable per declaration line:

StackObjSet LargeArrayObjs;
StackObjSet SmallArrayObjs;
StackObjSet AddrOfObjs;

lib/CodeGen/PrologEpilogInserter.cpp

547 ↗

(On Diff #6440)

Is this code identical to the one in CodeGen/LocalStackSlotAllocation.cpp? If so, please refactor it.

557 ↗

(On Diff #6440)

Same comment here as above. One variable per declaration please.

[jmagee]

See my inline comment in response to the refactoring. All the rest of the comments look good - I'll make those changes.

lib/CodeGen/PrologEpilogInserter.cpp

547 ↗

(On Diff #6440)

A large amount of the code in LocalStackSlotAllocation.cpp is identical to PrologEpilogInserter.cpp - there is significant refactoring that can be done there. It is not easy to draw the line...

I'm happy to refactor the stack-protector stuff out as a first step. I'm also happy to continue that effort to refactor out much of the duplication in these passes - the duplication always puzzled (and bothered) me a bit.

I'd prefer to do the refactoring as subsequent patches. Does that sound reasonable?

[void]

LGTM with above discussed changes.

lib/CodeGen/PrologEpilogInserter.cpp

547 ↗

(On Diff #6440)

A subsequent patch is fine, it's not like this is your fault or something. :-) I think that any significant refactoring you can do here would be an overall benefit. Thanks!

[jmagee]

Closed by commit rL200601 (authored by @jmagee).

[jmagee]

Thanks for the review! I'll get cracking on those PEI/LocalStackSlot refactors.

llvm/trunk/docs/LangRef.rst

``ssp``
     heuristic is used to determine if a function needs stack protectors
     or not. The heuristic used will enable protectors for functions with:
 
     - Character arrays larger than ``ssp-buffer-size`` (default 8).
     - Aggregates containing character arrays larger than ``ssp-buffer-size``.
     - Calls to alloca() with variable sizes or constant sizes greater than
       ``ssp-buffer-size``.
 
+    Variables that are identified as requiring a protector will be arranged
+    on the stack such that they are adjacent to the stack protector guard.
+
     If a function that has an ``ssp`` attribute is inlined into a
     function that doesn't have an ``ssp`` attribute, then the resulting
     function will have an ``ssp`` attribute.
 ``sspreq``
     This attribute indicates that the function should *always* emit a
     stack smashing protector. This overrides the ``ssp`` function
     attribute.
 
+    Variables that are identified as requiring a protector will be arranged
+    on the stack such that they are adjacent to the stack protector guard.
+    The specific layout rules are:
+
+    #. Large arrays and structures containing large arrays
+       (``>= ssp-buffer-size``) are closest to the stack protector.
+    #. Small arrays and structures containing small arrays
+       (``< ssp-buffer-size``) are 2nd closest to the protector.
+    #. Variables that have had their address taken are 3rd closest to the
+       protector.
+
     If a function that has an ``sspreq`` attribute is inlined into a
     function that doesn't have an ``sspreq`` attribute or which has an
     ``ssp`` or ``sspstrong`` attribute, then the resulting function will have
     an ``sspreq`` attribute.
 ``sspstrong``
     This attribute indicates that the function should emit a stack smashing
     protector. This attribute causes a strong heuristic to be used when
     determining if a function needs stack protectors.  The strong heuristic
     will enable protectors for functions with:
 
     - Arrays of any size and type
     - Aggregates containing an array of any size and type.
     - Calls to alloca().
     - Local variables that have had their address taken.
 
+    Variables that are identified as requiring a protector will be arranged
+    on the stack such that they are adjacent to the stack protector guard.
+    The specific layout rules are:
+
+    #. Large arrays and structures containing large arrays
+       (``>= ssp-buffer-size``) are closest to the stack protector.
+    #. Small arrays and structures containing small arrays
+       (``< ssp-buffer-size``) are 2nd closest to the protector.
+    #. Variables that have had their address taken are 3rd closest to the
+       protector.
+
     This overrides the ``ssp`` function attribute.
 
     If a function that has an ``sspstrong`` attribute is inlined into a
     function that doesn't have an ``sspstrong`` attribute, then the
     resulting function will have an ``sspstrong`` attribute.
 ``uwtable``
     This attribute indicates that the ABI being targeted requires that
     an unwind table entry be produce for this function even if we can

llvm/trunk/lib/CodeGen/LocalStackSlotAllocation.cpp

void LocalStackSlotPass::calculateFrameObjectOffsets(MachineFunction &Fn) {
   unsigned MaxAlign = 0;
   StackProtector *SP = &getAnalysis<StackProtector>();
 
   // Make sure that the stack protector comes before the local variables on the
   // stack.
   SmallSet<int, 16> ProtectedObjs;
   if (MFI->getStackProtectorIndex() >= 0) {
     StackObjSet LargeArrayObjs;
+    StackObjSet SmallArrayObjs;
+    StackObjSet AddrOfObjs;
+
     AdjustStackOffset(MFI, MFI->getStackProtectorIndex(), Offset,
                       StackGrowsDown, MaxAlign);
 
     // Assign large stack objects first.
     for (unsigned i = 0, e = MFI->getObjectIndexEnd(); i != e; ++i) {
       if (MFI->isDeadObjectIndex(i))
         continue;
       if (MFI->getStackProtectorIndex() == (int)i)
         continue;
 
       switch (SP->getSSPLayout(MFI->getObjectAllocation(i))) {
       case StackProtector::SSPLK_None:
+        continue;
       case StackProtector::SSPLK_SmallArray:
+        SmallArrayObjs.insert(i);
+        continue;
       case StackProtector::SSPLK_AddrOf:
+        AddrOfObjs.insert(i);
         continue;
       case StackProtector::SSPLK_LargeArray:
         LargeArrayObjs.insert(i);
         continue;
       }
       llvm_unreachable("Unexpected SSPLayoutKind.");
     }
 
     AssignProtectedObjSet(LargeArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
                           Offset, MaxAlign);
+    AssignProtectedObjSet(SmallArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
+                          Offset, MaxAlign);
+    AssignProtectedObjSet(AddrOfObjs, ProtectedObjs, MFI, StackGrowsDown,
+                          Offset, MaxAlign);
   }
 
   // Then assign frame offsets to stack objects that are not used to spill
   // callee saved registers.
   for (unsigned i = 0, e = MFI->getObjectIndexEnd(); i != e; ++i) {
     if (MFI->isDeadObjectIndex(i))
       continue;
     if (MFI->getStackProtectorIndex() == (int)i)

llvm/trunk/lib/CodeGen/PrologEpilogInserter.cpp

if (MFI->getUseLocalStackAllocationBlock()) {
     MaxAlign = std::max(Align, MaxAlign);
   }
 
   // Make sure that the stack protector comes before the local variables on the
   // stack.
   SmallSet<int, 16> ProtectedObjs;
   if (MFI->getStackProtectorIndex() >= 0) {
     StackObjSet LargeArrayObjs;
+    StackObjSet SmallArrayObjs;
+    StackObjSet AddrOfObjs;
+
     AdjustStackOffset(MFI, MFI->getStackProtectorIndex(), StackGrowsDown,
                       Offset, MaxAlign);
 
     // Assign large stack objects first.
     for (unsigned i = 0, e = MFI->getObjectIndexEnd(); i != e; ++i) {
       if (MFI->isObjectPreAllocated(i) &&
           MFI->getUseLocalStackAllocationBlock())
         continue;
       if (i >= MinCSFrameIndex && i <= MaxCSFrameIndex)
         continue;
       if (RS && RS->isScavengingFrameIndex((int)i))
         continue;
       if (MFI->isDeadObjectIndex(i))
         continue;
       if (MFI->getStackProtectorIndex() == (int)i)
         continue;
 
       switch (SP->getSSPLayout(MFI->getObjectAllocation(i))) {
       case StackProtector::SSPLK_None:
+        continue;
       case StackProtector::SSPLK_SmallArray:
+        SmallArrayObjs.insert(i);
+        continue;
       case StackProtector::SSPLK_AddrOf:
+        AddrOfObjs.insert(i);
         continue;
       case StackProtector::SSPLK_LargeArray:
         LargeArrayObjs.insert(i);
         continue;
       }
       llvm_unreachable("Unexpected SSPLayoutKind.");
     }
 
     AssignProtectedObjSet(LargeArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
                           Offset, MaxAlign);
+    AssignProtectedObjSet(SmallArrayObjs, ProtectedObjs, MFI, StackGrowsDown,
+                          Offset, MaxAlign);
+    AssignProtectedObjSet(AddrOfObjs, ProtectedObjs, MFI, StackGrowsDown,
+                          Offset, MaxAlign);
   }
 
   // Then assign frame offsets to stack objects that are not used to spill
   // callee saved registers.
   for (unsigned i = 0, e = MFI->getObjectIndexEnd(); i != e; ++i) {
     if (MFI->isObjectPreAllocated(i) &&
         MFI->getUseLocalStackAllocationBlock())
       continue;

llvm/trunk/test/CodeGen/ARM/ssp-data-layout.ll

; CHECK: bl end_struct_small_nonchar
   %6 = load i16* %5, align 1
   %coerce.dive26 = getelementptr %struct.struct_small_nonchar* %d, i32 0, i32 0
   %7 = bitcast [2 x i16]* %coerce.dive26 to i32*
   %8 = load i32* %7, align 1
   call void @takes_all(i64 %4, i16 %6, %struct.struct_large_nonchar* byval align 8 %c, i32 %8, i8* %arraydecay, i8* %arraydecay22, i32* %arraydecay23, i16* %arraydecay24, i32* %ptr, i32 %0, i32 %1, i32 %2)
   ret void
 }
 
+define void @layout_sspstrong() sspstrong {
+entry:
+; Expected stack layout for sspstrong is
+; 144  large_nonchar          . Group 1, nested arrays,
+; 136  large_char             .  arrays >= ssp-buffer-size
+; 128  struct_large_char      .
+; 96   struct_large_nonchar   .
+; 84+8 small_non_char         | Group 2, nested arrays, 
+; 90   small_char             |  arrays < ssp-buffer-size
+; 88   struct_small_char      |
+; 84   struct_small_nonchar   |
+; 80   addrof                 * Group 3, addr-of local
+; 76   scalar1                + Group 4, everything else
+; 72   scalar2                +
+; 68   scalar3                +
+;   
+; CHECK: layout_sspstrong:
+; r[[SP]] is used as an offset into the stack later
+; CHECK: add r[[SP:[0-9]+]], sp, #84
+
+; CHECK: bl get_scalar1
+; CHECK: str r0, [sp, #76]
+; CHECK: bl end_scalar1
+
+; CHECK: bl get_scalar2
+; CHECK: str r0, [sp, #72]
+; CHECK: bl end_scalar2
+
+; CHECK: bl get_scalar3
+; CHECK: str r0, [sp, #68]
+; CHECK: bl end_scalar3
+
+; CHECK: bl get_addrof
+; CHECK: str r0, [sp, #80]
+; CHECK: bl end_addrof
+
+; CHECK: get_small_nonchar
+; CHECK: strh r0, [r[[SP]], #8]
+; CHECK: bl end_small_nonchar
+
+; CHECK: bl get_large_nonchar
+; CHECK: str r0, [sp, #144]
+; CHECK: bl end_large_nonchar
+
+; CHECK: bl get_small_char
+; CHECK: strb r0, [sp, #90]
+; CHECK: bl end_small_char
+
+; CHECK: bl get_large_char
+; CHECK: strb r0, [sp, #136]
+; CHECK: bl end_large_char
+
+; CHECK: bl get_struct_large_char
+; CHECK: strb r0, [sp, #128]
+; CHECK: bl end_struct_large_char
+
+; CHECK: bl get_struct_small_char
+; CHECK: strb r0, [sp, #88]
+; CHECK: bl end_struct_small_char
+
+; CHECK: bl get_struct_large_nonchar
+; CHECK: str r0, [sp, #96]
+; CHECK: bl end_struct_large_nonchar
+
+; CHECK: bl get_struct_small_nonchar
+; CHECK: strh r0, [r[[SP]]]
+; CHECK: bl end_struct_small_nonchar
+  %x = alloca i32, align 4
+  %y = alloca i32, align 4
+  %z = alloca i32, align 4
+  %ptr = alloca i32, align 4
+  %small2 = alloca [2 x i16], align 2
+  %large2 = alloca [8 x i32], align 16
+  %small = alloca [2 x i8], align 1
+  %large = alloca [8 x i8], align 1
+  %a = alloca %struct.struct_large_char, align 1
+  %b = alloca %struct.struct_small_char, align 1
+  %c = alloca %struct.struct_large_nonchar, align 8
+  %d = alloca %struct.struct_small_nonchar, align 2
+  %call = call i32 @get_scalar1()
+  store i32 %call, i32* %x, align 4
+  call void @end_scalar1()
+  %call1 = call i32 @get_scalar2()
+  store i32 %call1, i32* %y, align 4
+  call void @end_scalar2()
+  %call2 = call i32 @get_scalar3()
+  store i32 %call2, i32* %z, align 4
+  call void @end_scalar3()
+  %call3 = call i32 @get_addrof()
+  store i32 %call3, i32* %ptr, align 4
+  call void @end_addrof()
+  %call4 = call signext i16 @get_small_nonchar()
+  %arrayidx = getelementptr inbounds [2 x i16]* %small2, i32 0, i64 0
+  store i16 %call4, i16* %arrayidx, align 2
+  call void @end_small_nonchar()
+  %call5 = call i32 @get_large_nonchar()
+  %arrayidx6 = getelementptr inbounds [8 x i32]* %large2, i32 0, i64 0
+  store i32 %call5, i32* %arrayidx6, align 4
+  call void @end_large_nonchar()
+  %call7 = call signext i8 @get_small_char()
+  %arrayidx8 = getelementptr inbounds [2 x i8]* %small, i32 0, i64 0
+  store i8 %call7, i8* %arrayidx8, align 1
+  call void @end_small_char()
+  %call9 = call signext i8 @get_large_char()
+  %arrayidx10 = getelementptr inbounds [8 x i8]* %large, i32 0, i64 0
+  store i8 %call9, i8* %arrayidx10, align 1
+  call void @end_large_char()
+  %call11 = call signext i8 @get_struct_large_char()
+  %foo = getelementptr inbounds %struct.struct_large_char* %a, i32 0, i32 0
+  %arrayidx12 = getelementptr inbounds [8 x i8]* %foo, i32 0, i64 0
+  store i8 %call11, i8* %arrayidx12, align 1
+  call void @end_struct_large_char()
+  %call13 = call signext i8 @get_struct_small_char()
+  %foo14 = getelementptr inbounds %struct.struct_small_char* %b, i32 0, i32 0
+  %arrayidx15 = getelementptr inbounds [2 x i8]* %foo14, i32 0, i64 0
+  store i8 %call13, i8* %arrayidx15, align 1
+  call void @end_struct_small_char()
+  %call16 = call i32 @get_struct_large_nonchar()
+  %foo17 = getelementptr inbounds %struct.struct_large_nonchar* %c, i32 0, i32 0
+  %arrayidx18 = getelementptr inbounds [8 x i32]* %foo17, i32 0, i64 0
+  store i32 %call16, i32* %arrayidx18, align 4
+  call void @end_struct_large_nonchar()
+  %call19 = call signext i16 @get_struct_small_nonchar()
+  %foo20 = getelementptr inbounds %struct.struct_small_nonchar* %d, i32 0, i32 0
+  %arrayidx21 = getelementptr inbounds [2 x i16]* %foo20, i32 0, i64 0
+  store i16 %call19, i16* %arrayidx21, align 2
+  call void @end_struct_small_nonchar()
+  %arraydecay = getelementptr inbounds [8 x i8]* %large, i32 0, i32 0
+  %arraydecay22 = getelementptr inbounds [2 x i8]* %small, i32 0, i32 0
+  %arraydecay23 = getelementptr inbounds [8 x i32]* %large2, i32 0, i32 0
+  %arraydecay24 = getelementptr inbounds [2 x i16]* %small2, i32 0, i32 0
+  %0 = load i32* %x, align 4
+  %1 = load i32* %y, align 4
+  %2 = load i32* %z, align 4
+  %coerce.dive = getelementptr %struct.struct_large_char* %a, i32 0, i32 0
+  %3 = bitcast [8 x i8]* %coerce.dive to i64*
+  %4 = load i64* %3, align 1
+  %coerce.dive25 = getelementptr %struct.struct_small_char* %b, i32 0, i32 0
+  %5 = bitcast [2 x i8]* %coerce.dive25 to i16*
+  %6 = load i16* %5, align 1
+  %coerce.dive26 = getelementptr %struct.struct_small_nonchar* %d, i32 0, i32 0
+  %7 = bitcast [2 x i16]* %coerce.dive26 to i32*
+  %8 = load i32* %7, align 1
+  call void @takes_all(i64 %4, i16 %6, %struct.struct_large_nonchar* byval align 8 %c, i32 %8, i8* %arraydecay, i8* %arraydecay22, i32* %arraydecay23, i16* %arraydecay24, i32* %ptr, i32 %0, i32 %1, i32 %2)
+  ret void
+}
+
+define void @layout_sspreq() sspreq {
+entry:
+; Expected stack layout for sspreq is the same as sspstrong
+;   
+; CHECK: layout_sspreq:
+; r[[SP]] is used as an offset into the stack later
+; CHECK: add r[[SP:[0-9]+]], sp, #84
+
+; CHECK: bl get_scalar1
+; CHECK: str r0, [sp, #76]
+; CHECK: bl end_scalar1
+
+; CHECK: bl get_scalar2
+; CHECK: str r0, [sp, #72]
+; CHECK: bl end_scalar2
+
+; CHECK: bl get_scalar3
+; CHECK: str r0, [sp, #68]
+; CHECK: bl end_scalar3
+
+; CHECK: bl get_addrof
+; CHECK: str r0, [sp, #80]
+; CHECK: bl end_addrof
+
+; CHECK: get_small_nonchar
+; CHECK: strh r0, [r[[SP]], #8]
+; CHECK: bl end_small_nonchar
+
+; CHECK: bl get_large_nonchar
+; CHECK: str r0, [sp, #144]
+; CHECK: bl end_large_nonchar
+
+; CHECK: bl get_small_char
+; CHECK: strb r0, [sp, #90]
+; CHECK: bl end_small_char
+
+; CHECK: bl get_large_char
+; CHECK: strb r0, [sp, #136]
+; CHECK: bl end_large_char
+
+; CHECK: bl get_struct_large_char
+; CHECK: strb r0, [sp, #128]
+; CHECK: bl end_struct_large_char
+
+; CHECK: bl get_struct_small_char
+; CHECK: strb r0, [sp, #88]
+; CHECK: bl end_struct_small_char
+
+; CHECK: bl get_struct_large_nonchar
+; CHECK: str r0, [sp, #96]
+; CHECK: bl end_struct_large_nonchar
+
+; CHECK: bl get_struct_small_nonchar
+; CHECK: strh r0, [r[[SP]]]
+; CHECK: bl end_struct_small_nonchar
+  %x = alloca i32, align 4
+  %y = alloca i32, align 4
+  %z = alloca i32, align 4
+  %ptr = alloca i32, align 4
+  %small2 = alloca [2 x i16], align 2
+  %large2 = alloca [8 x i32], align 16
+  %small = alloca [2 x i8], align 1
+  %large = alloca [8 x i8], align 1
+  %a = alloca %struct.struct_large_char, align 1
+  %b = alloca %struct.struct_small_char, align 1
+  %c = alloca %struct.struct_large_nonchar, align 8
+  %d = alloca %struct.struct_small_nonchar, align 2
+  %call = call i32 @get_scalar1()
+  store i32 %call, i32* %x, align 4
+  call void @end_scalar1()
+  %call1 = call i32 @get_scalar2()
+  store i32 %call1, i32* %y, align 4
+  call void @end_scalar2()
+  %call2 = call i32 @get_scalar3()
+  store i32 %call2, i32* %z, align 4
+  call void @end_scalar3()
+  %call3 = call i32 @get_addrof()
+  store i32 %call3, i32* %ptr, align 4
+  call void @end_addrof()
+  %call4 = call signext i16 @get_small_nonchar()
+  %arrayidx = getelementptr inbounds [2 x i16]* %small2, i32 0, i64 0
+  store i16 %call4, i16* %arrayidx, align 2
+  call void @end_small_nonchar()
+  %call5 = call i32 @get_large_nonchar()
+  %arrayidx6 = getelementptr inbounds [8 x i32]* %large2, i32 0, i64 0
+  store i32 %call5, i32* %arrayidx6, align 4
+  call void @end_large_nonchar()
+  %call7 = call signext i8 @get_small_char()
+  %arrayidx8 = getelementptr inbounds [2 x i8]* %small, i32 0, i64 0
+  store i8 %call7, i8* %arrayidx8, align 1
+  call void @end_small_char()
+  %call9 = call signext i8 @get_large_char()
+  %arrayidx10 = getelementptr inbounds [8 x i8]* %large, i32 0, i64 0
+  store i8 %call9, i8* %arrayidx10, align 1
+  call void @end_large_char()
+  %call11 = call signext i8 @get_struct_large_char()
+  %foo = getelementptr inbounds %struct.struct_large_char* %a, i32 0, i32 0
+  %arrayidx12 = getelementptr inbounds [8 x i8]* %foo, i32 0, i64 0
+  store i8 %call11, i8* %arrayidx12, align 1
+  call void @end_struct_large_char()
+  %call13 = call signext i8 @get_struct_small_char()
+  %foo14 = getelementptr inbounds %struct.struct_small_char* %b, i32 0, i32 0
+  %arrayidx15 = getelementptr inbounds [2 x i8]* %foo14, i32 0, i64 0
+  store i8 %call13, i8* %arrayidx15, align 1
+  call void @end_struct_small_char()
+  %call16 = call i32 @get_struct_large_nonchar()
+  %foo17 = getelementptr inbounds %struct.struct_large_nonchar* %c, i32 0, i32 0
+  %arrayidx18 = getelementptr inbounds [8 x i32]* %foo17, i32 0, i64 0
+  store i32 %call16, i32* %arrayidx18, align 4
+  call void @end_struct_large_nonchar()
+  %call19 = call signext i16 @get_struct_small_nonchar()
+  %foo20 = getelementptr inbounds %struct.struct_small_nonchar* %d, i32 0, i32 0
+  %arrayidx21 = getelementptr inbounds [2 x i16]* %foo20, i32 0, i64 0
+  store i16 %call19, i16* %arrayidx21, align 2
+  call void @end_struct_small_nonchar()
+  %arraydecay = getelementptr inbounds [8 x i8]* %large, i32 0, i32 0
+  %arraydecay22 = getelementptr inbounds [2 x i8]* %small, i32 0, i32 0
+  %arraydecay23 = getelementptr inbounds [8 x i32]* %large2, i32 0, i32 0
+  %arraydecay24 = getelementptr inbounds [2 x i16]* %small2, i32 0, i32 0
+  %0 = load i32* %x, align 4
+  %1 = load i32* %y, align 4
+  %2 = load i32* %z, align 4
+  %coerce.dive = getelementptr %struct.struct_large_char* %a, i32 0, i32 0
+  %3 = bitcast [8 x i8]* %coerce.dive to i64*
+  %4 = load i64* %3, align 1
+  %coerce.dive25 = getelementptr %struct.struct_small_char* %b, i32 0, i32 0
+  %5 = bitcast [2 x i8]* %coerce.dive25 to i16*
+  %6 = load i16* %5, align 1
+  %coerce.dive26 = getelementptr %struct.struct_small_nonchar* %d, i32 0, i32 0
+  %7 = bitcast [2 x i16]* %coerce.dive26 to i32*
+  %8 = load i32* %7, align 1
+  call void @takes_all(i64 %4, i16 %6, %struct.struct_large_nonchar* byval align 8 %c, i32 %8, i8* %arraydecay, i8* %arraydecay22, i32* %arraydecay23, i16* %arraydecay24, i32* %ptr, i32 %0, i32 %1, i32 %2)
+  ret void
+}
+
+define void @struct_with_protectable_arrays() sspstrong {
+entry:
+; Check to ensure that a structure which contains a small array followed by a
+; large array is assigned to the stack properly as a large object.
+; CHECK: struct_with_protectable_arrays:
+; CHECK: bl get_struct_small_char
+; CHECK: strb r0, [sp, #68]
+; CHECK: bl end_struct_small_char
+; CHECK: bl get_struct_large_char2
+; CHECK: strb r0, [sp, #106]
+; CHECK: bl end_struct_large_char2
+  %a = alloca %struct.struct_small_char, align 1
+  %b = alloca %struct.struct_large_char2, align 1
+  %d1 = alloca %struct.struct_large_nonchar, align 8
+  %d2 = alloca %struct.struct_small_nonchar, align 2
+  %call = call signext i8 @get_struct_small_char()
+  %foo = getelementptr inbounds %struct.struct_small_char* %a, i32 0, i32 0
+  %arrayidx = getelementptr inbounds [2 x i8]* %foo, i32 0, i64 0
+  store i8 %call, i8* %arrayidx, align 1
+  call void @end_struct_small_char()
+  %call1 = call signext i8 @get_struct_large_char2()
+  %foo2 = getelementptr inbounds %struct.struct_large_char2* %b, i32 0, i32 1
+  %arrayidx3 = getelementptr inbounds [8 x i8]* %foo2, i32 0, i64 0
+  store i8 %call1, i8* %arrayidx3, align 1
+  call void @end_struct_large_char2()
+  %0 = bitcast %struct.struct_large_char2* %b to %struct.struct_large_char*
+  %coerce.dive = getelementptr %struct.struct_large_char* %0, i32 0, i32 0
+  %1 = bitcast [8 x i8]* %coerce.dive to i64*
+  %2 = load i64* %1, align 1
+  %coerce.dive4 = getelementptr %struct.struct_small_char* %a, i32 0, i32 0
+  %3 = bitcast [2 x i8]* %coerce.dive4 to i16*
+  %4 = load i16* %3, align 1
+  %coerce.dive5 = getelementptr %struct.struct_small_nonchar* %d2, i32 0, i32 0
+  %5 = bitcast [2 x i16]* %coerce.dive5 to i32*
+  %6 = load i32* %5, align 1
+  call void @takes_all(i64 %2, i16 %4, %struct.struct_large_nonchar* byval align 8 %d1, i32 %6, i8* null, i8* null, i32* null, i16* null, i32* null, i32 0, i32 0, i32 0)
+  ret void
+}
+
 declare i32 @get_scalar1()
 declare void @end_scalar1()
 
 declare i32 @get_scalar2()
 declare void @end_scalar2()
 
 declare i32 @get_scalar3()
 declare void @end_scalar3()

llvm/trunk/test/CodeGen/X86/ssp-data-layout.ll

 ; FastISel was not setting the StackProtectorIndex when lowering
 ; Intrinsic::stackprotector and as a result the stack re-arrangement code was
 ; never applied.  This problem only shows up on non-Linux platforms because on
 ; Linux the stack protector cookie is loaded from a special address space which
 ; always triggers standard ISel.  Run a basic test to ensure that at -O0
 ; on a non-linux target the data layout rules are triggered.
 
 %struct.struct_large_char = type { [8 x i8] }
-%struct.struct_large_char2 = type { [2 x i8], [8 x i8] }
 %struct.struct_small_char = type { [2 x i8] }
 %struct.struct_large_nonchar = type { [8 x i32] }
 %struct.struct_small_nonchar = type { [2 x i16] }
 
 define void @layout_ssp() ssp {
 entry:
 ; Expected stack layout for ssp is
 ;  -16 large_char          . Group 1, nested arrays, arrays >= ssp-buffer-size
; CHECK: call{{l|q}} end_struct_small_nonchar
   %6 = load i16* %5, align 1
   %coerce.dive26 = getelementptr %struct.struct_small_nonchar* %d, i32 0, i32 0
   %7 = bitcast [2 x i16]* %coerce.dive26 to i32*
   %8 = load i32* %7, align 1
   call void @takes_all(i64 %4, i16 %6, %struct.struct_large_nonchar* byval align 8 %c, i32 %8, i8* %arraydecay, i8* %arraydecay22, i32* %arraydecay23, i16* %arraydecay24, i32* %ptr, i32 %0, i32 %1, i32 %2)
   ret void
 }
 
+define void @layout_sspstrong() nounwind uwtable sspstrong {
+entry:
+; Expected stack layout for sspstrong is
+;   -48   large_nonchar          . Group 1, nested arrays,
+;   -56   large_char             .  arrays >= ssp-buffer-size
+;   -64   struct_large_char      .
+;   -96   struct_large_nonchar   .
+;   -100  small_non_char         | Group 2, nested arrays, 
+;   -102  small_char             |  arrays < ssp-buffer-size
+;   -104  struct_small_char      |
+;   -112  struct_small_nonchar   |
+;   -116  addrof                 * Group 3, addr-of local
+;   -120  scalar                 + Group 4, everything else
+;   -124  scalar                 +
+;   -128  scalar                 +
+;   
+; CHECK: layout_sspstrong:
+; CHECK: call{{l|q}} get_scalar1
+; CHECK: movl %eax, -120(
+; CHECK: call{{l|q}} end_scalar1
+
+; CHECK: call{{l|q}} get_scalar2
+; CHECK: movl %eax, -124(
+; CHECK: call{{l|q}} end_scalar2
+
+; CHECK: call{{l|q}} get_scalar3
+; CHECK: movl %eax, -128(
+; CHECK: call{{l|q}} end_scalar3
+
+; CHECK: call{{l|q}} get_addrof
+; CHECK: movl %eax, -116(
+; CHECK: call{{l|q}} end_addrof
+
+; CHECK: get_small_nonchar
+; CHECK: movw %ax, -100(
+; CHECK: call{{l|q}} end_small_nonchar
+
+; CHECK: call{{l|q}} get_large_nonchar
+; CHECK: movl %eax, -48(
+; CHECK: call{{l|q}} end_large_nonchar
+
+; CHECK: call{{l|q}} get_small_char
+; CHECK: movb %al, -102(
+; CHECK: call{{l|q}} end_small_char
+
+; CHECK: call{{l|q}} get_large_char
+; CHECK: movb %al, -56(
+; CHECK: call{{l|q}} end_large_char
+
+; CHECK: call{{l|q}} get_struct_large_char
+; CHECK: movb %al, -64(
+; CHECK: call{{l|q}} end_struct_large_char
+
+; CHECK: call{{l|q}} get_struct_small_char
+; CHECK: movb %al, -104(
+; CHECK: call{{l|q}} end_struct_small_char
+
+; CHECK: call{{l|q}} get_struct_large_nonchar
+; CHECK: movl %eax, -96(
+; CHECK: call{{l|q}} end_struct_large_nonchar
+
+; CHECK: call{{l|q}} get_struct_small_nonchar
+; CHECK: movw %ax, -112(
+; CHECK: call{{l|q}} end_struct_small_nonchar
+  %x = alloca i32, align 4
+  %y = alloca i32, align 4
+  %z = alloca i32, align 4
+  %ptr = alloca i32, align 4
+  %small2 = alloca [2 x i16], align 2
+  %large2 = alloca [8 x i32], align 16
+  %small = alloca [2 x i8], align 1
+  %large = alloca [8 x i8], align 1
+  %a = alloca %struct.struct_large_char, align 1
+  %b = alloca %struct.struct_small_char, align 1
+  %c = alloca %struct.struct_large_nonchar, align 8
+  %d = alloca %struct.struct_small_nonchar, align 2
+  %call = call i32 @get_scalar1()
+  store i32 %call, i32* %x, align 4
+  call void @end_scalar1()
+  %call1 = call i32 @get_scalar2()
+  store i32 %call1, i32* %y, align 4
+  call void @end_scalar2()
+  %call2 = call i32 @get_scalar3()
+  store i32 %call2, i32* %z, align 4
+  call void @end_scalar3()
+  %call3 = call i32 @get_addrof()
+  store i32 %call3, i32* %ptr, align 4
+  call void @end_addrof()
+  %call4 = call signext i16 @get_small_nonchar()
+  %arrayidx = getelementptr inbounds [2 x i16]* %small2, i32 0, i64 0
+  store i16 %call4, i16* %arrayidx, align 2
+  call void @end_small_nonchar()
+  %call5 = call i32 @get_large_nonchar()
+  %arrayidx6 = getelementptr inbounds [8 x i32]* %large2, i32 0, i64 0
+  store i32 %call5, i32* %arrayidx6, align 4
+  call void @end_large_nonchar()
+  %call7 = call signext i8 @get_small_char()
+  %arrayidx8 = getelementptr inbounds [2 x i8]* %small, i32 0, i64 0
+  store i8 %call7, i8* %arrayidx8, align 1
+  call void @end_small_char()
+  %call9 = call signext i8 @get_large_char()
+  %arrayidx10 = getelementptr inbounds [8 x i8]* %large, i32 0, i64 0
+  store i8 %call9, i8* %arrayidx10, align 1
+  call void @end_large_char()
+  %call11 = call signext i8 @get_struct_large_char()
+  %foo = getelementptr inbounds %struct.struct_large_char* %a, i32 0, i32 0
+  %arrayidx12 = getelementptr inbounds [8 x i8]* %foo, i32 0, i64 0
+  store i8 %call11, i8* %arrayidx12, align 1
+  call void @end_struct_large_char()
+  %call13 = call signext i8 @get_struct_small_char()
+  %foo14 = getelementptr inbounds %struct.struct_small_char* %b, i32 0, i32 0
+  %arrayidx15 = getelementptr inbounds [2 x i8]* %foo14, i32 0, i64 0
+  store i8 %call13, i8* %arrayidx15, align 1
+  call void @end_struct_small_char()
+  %call16 = call i32 @get_struct_large_nonchar()
+  %foo17 = getelementptr inbounds %struct.struct_large_nonchar* %c, i32 0, i32 0
+  %arrayidx18 = getelementptr inbounds [8 x i32]* %foo17, i32 0, i64 0
+  store i32 %call16, i32* %arrayidx18, align 4
+  call void @end_struct_large_nonchar()
+  %call19 = call signext i16 @get_struct_small_nonchar()
+  %foo20 = getelementptr inbounds %struct.struct_small_nonchar* %d, i32 0, i32 0
+  %arrayidx21 = getelementptr inbounds [2 x i16]* %foo20, i32 0, i64 0
+  store i16 %call19, i16* %arrayidx21, align 2
+  call void @end_struct_small_nonchar()
+  %arraydecay = getelementptr inbounds [8 x i8]* %large, i32 0, i32 0
+  %arraydecay22 = getelementptr inbounds [2 x i8]* %small, i32 0, i32 0
+  %arraydecay23 = getelementptr inbounds [8 x i32]* %large2, i32 0, i32 0
+  %arraydecay24 = getelementptr inbounds [2 x i16]* %small2, i32 0, i32 0
+  %0 = load i32* %x, align 4
+  %1 = load i32* %y, align 4
+  %2 = load i32* %z, align 4
+  %coerce.dive = getelementptr %struct.struct_large_char* %a, i32 0, i32 0
+  %3 = bitcast [8 x i8]* %coerce.dive to i64*
+  %4 = load i64* %3, align 1
+  %coerce.dive25 = getelementptr %struct.struct_small_char* %b, i32 0, i32 0
+  %5 = bitcast [2 x i8]* %coerce.dive25 to i16*
+  %6 = load i16* %5, align 1
+  %coerce.dive26 = getelementptr %struct.struct_small_nonchar* %d, i32 0, i32 0
+  %7 = bitcast [2 x i16]* %coerce.dive26 to i32*
+  %8 = load i32* %7, align 1
+  call void @takes_all(i64 %4, i16 %6, %struct.struct_large_nonchar* byval align 8 %c, i32 %8, i8* %arraydecay, i8* %arraydecay22, i32* %arraydecay23, i16* %arraydecay24, i32* %ptr, i32 %0, i32 %1, i32 %2)
+  ret void
+}
+
+define void @layout_sspreq() nounwind uwtable sspreq {
+entry:
+; Expected stack layout for sspreq is the same as sspstrong
+;   
+; CHECK: layout_sspreq:
+; CHECK: call{{l|q}} get_scalar1
+; CHECK: movl %eax, -120(
+; CHECK: call{{l|q}} end_scalar1
+
+; CHECK: call{{l|q}} get_scalar2
+; CHECK: movl %eax, -124(
+; CHECK: call{{l|q}} end_scalar2
+
+; CHECK: call{{l|q}} get_scalar3
+; CHECK: movl %eax, -128(
+; CHECK: call{{l|q}} end_scalar3
+
+; CHECK: call{{l|q}} get_addrof
+; CHECK: movl %eax, -116(
+; CHECK: call{{l|q}} end_addrof
+
+; CHECK: get_small_nonchar
+; CHECK: movw %ax, -100(
+; CHECK: call{{l|q}} end_small_nonchar
+
+; CHECK: call{{l|q}} get_large_nonchar
+; CHECK: movl %eax, -48(
+; CHECK: call{{l|q}} end_large_nonchar
+
+; CHECK: call{{l|q}} get_small_char
+; CHECK: movb %al, -102(
+; CHECK: call{{l|q}} end_small_char
+
+; CHECK: call{{l|q}} get_large_char
+; CHECK: movb %al, -56(
+; CHECK: call{{l|q}} end_large_char
+
+; CHECK: call{{l|q}} get_struct_large_char
+; CHECK: movb %al, -64(
+; CHECK: call{{l|q}} end_struct_large_char
+
+; CHECK: call{{l|q}} get_struct_small_char
+; CHECK: movb %al, -104(
+; CHECK: call{{l|q}} end_struct_small_char
+
+; CHECK: call{{l|q}} get_struct_large_nonchar
+; CHECK: movl %eax, -96(
+; CHECK: call{{l|q}} end_struct_large_nonchar
+
+; CHECK: call{{l|q}} get_struct_small_nonchar
+; CHECK: movw %ax, -112(
+; CHECK: call{{l|q}} end_struct_small_nonchar
+  %x = alloca i32, align 4
+  %y = alloca i32, align 4
+  %z = alloca i32, align 4
+  %ptr = alloca i32, align 4
+  %small2 = alloca [2 x i16], align 2
+  %large2 = alloca [8 x i32], align 16
+  %small = alloca [2 x i8], align 1
+  %large = alloca [8 x i8], align 1
+  %a = alloca %struct.struct_large_char, align 1
+  %b = alloca %struct.struct_small_char, align 1
+  %c = alloca %struct.struct_large_nonchar, align 8
+  %d = alloca %struct.struct_small_nonchar, align 2
+  %call = call i32 @get_scalar1()
+  store i32 %call, i32* %x, align 4
+  call void @end_scalar1()
+  %call1 = call i32 @get_scalar2()
+  store i32 %call1, i32* %y, align 4
+  call void @end_scalar2()
+  %call2 = call i32 @get_scalar3()
+  store i32 %call2, i32* %z, align 4
+  call void @end_scalar3()
+  %call3 = call i32 @get_addrof()
+  store i32 %call3, i32* %ptr, align 4
+  call void @end_addrof()
+  %call4 = call signext i16 @get_small_nonchar()
+  %arrayidx = getelementptr inbounds [2 x i16]* %small2, i32 0, i64 0
+  store i16 %call4, i16* %arrayidx, align 2
+  call void @end_small_nonchar()
+  %call5 = call i32 @get_large_nonchar()
+  %arrayidx6 = getelementptr inbounds [8 x i32]* %large2, i32 0, i64 0
+  store i32 %call5, i32* %arrayidx6, align 4
+  call void @end_large_nonchar()
+  %call7 = call signext i8 @get_small_char()
+  %arrayidx8 = getelementptr inbounds [2 x i8]* %small, i32 0, i64 0
+  store i8 %call7, i8* %arrayidx8, align 1
+  call void @end_small_char()
+  %call9 = call signext i8 @get_large_char()
+  %arrayidx10 = getelementptr inbounds [8 x i8]* %large, i32 0, i64 0
+  store i8 %call9, i8* %arrayidx10, align 1
+  call void @end_large_char()
+  %call11 = call signext i8 @get_struct_large_char()
+  %foo = getelementptr inbounds %struct.struct_large_char* %a, i32 0, i32 0
+  %arrayidx12 = getelementptr inbounds [8 x i8]* %foo, i32 0, i64 0
+  store i8 %call11, i8* %arrayidx12, align 1
+  call void @end_struct_large_char()
+  %call13 = call signext i8 @get_struct_small_char()
+  %foo14 = getelementptr inbounds %struct.struct_small_char* %b, i32 0, i32 0
+  %arrayidx15 = getelementptr inbounds [2 x i8]* %foo14, i32 0, i64 0
+  store i8 %call13, i8* %arrayidx15, align 1
+  call void @end_struct_small_char()
+  %call16 = call i32 @get_struct_large_nonchar()
+  %foo17 = getelementptr inbounds %struct.struct_large_nonchar* %c, i32 0, i32 0
+  %arrayidx18 = getelementptr inbounds [8 x i32]* %foo17, i32 0, i64 0
+  store i32 %call16, i32* %arrayidx18, align 4
+  call void @end_struct_large_nonchar()
+  %call19 = call signext i16 @get_struct_small_nonchar()
+  %foo20 = getelementptr inbounds %struct.struct_small_nonchar* %d, i32 0, i32 0
+  %arrayidx21 = getelementptr inbounds [2 x i16]* %foo20, i32 0, i64 0
+  store i16 %call19, i16* %arrayidx21, align 2
+  call void @end_struct_small_nonchar()
+  %arraydecay = getelementptr inbounds [8 x i8]* %large, i32 0, i32 0
+  %arraydecay22 = getelementptr inbounds [2 x i8]* %small, i32 0, i32 0
+  %arraydecay23 = getelementptr inbounds [8 x i32]* %large2, i32 0, i32 0
+  %arraydecay24 = getelementptr inbounds [2 x i16]* %small2, i32 0, i32 0
+  %0 = load i32* %x, align 4
+  %1 = load i32* %y, align 4
+  %2 = load i32* %z, align 4
+  %coerce.dive = getelementptr %struct.struct_large_char* %a, i32 0, i32 0
+  %3 = bitcast [8 x i8]* %coerce.dive to i64*
+  %4 = load i64* %3, align 1
+  %coerce.dive25 = getelementptr %struct.struct_small_char* %b, i32 0, i32 0
+  %5 = bitcast [2 x i8]* %coerce.dive25 to i16*
+  %6 = load i16* %5, align 1
+  %coerce.dive26 = getelementptr %struct.struct_small_nonchar* %d, i32 0, i32 0
+  %7 = bitcast [2 x i16]* %coerce.dive26 to i32*
+  %8 = load i32* %7, align 1
+  call void @takes_all(i64 %4, i16 %6, %struct.struct_large_nonchar* byval align 8 %c, i32 %8, i8* %arraydecay, i8* %arraydecay22, i32* %arraydecay23, i16* %arraydecay24, i32* %ptr, i32 %0, i32 %1, i32 %2)
+  ret void
+}
+
 define void @fast_non_linux() ssp {
 entry:
 ; FAST-NON-LIN: fast_non_linux:
 ; FAST-NON-LIN: call{{l|q}} get_scalar1
 ; FAST-NON-LIN: movl %eax, -20(
 ; FAST-NON-LIN: call{{l|q}} end_scalar1
 
 ; FAST-NON-LIN: call{{l|q}} get_large_char

 declare void @end_small_char()
 
 declare signext i8 @get_large_char()
 declare void @end_large_char()
 
 declare signext i8 @get_struct_large_char()
 declare void @end_struct_large_char()
 
-declare signext i8 @get_struct_large_char2()
-declare void @end_struct_large_char2()
-
 declare signext i8 @get_struct_small_char()
 declare void @end_struct_small_char()
 
 declare i32 @get_struct_large_nonchar()
 declare void @end_struct_large_nonchar()
 
 declare signext i16 @get_struct_small_nonchar()
 declare void @end_struct_small_nonchar()
 
 declare void @takes_all(i64, i16, %struct.struct_large_nonchar* byval align 8, i32, i8*, i8*, i32*, i16*, i32*, i32, i32, i32)
 declare void @takes_two(i32, i8*)