[CodeGenPrepare] Fix use-after-free

If OptimizeExtractBits() encountered a shift instruction with no operands at all,
it would erase the instruction, but still return false.

This previously didn’t matter because its caller would always return after
processing the instruction, but https://reviews.llvm.org/D63233 changed the
function’s caller to fall through if it returned false, which would then cause
a use-after-free detectable by ASAN.

This change makes OptimizeExtractBits return true if it removes a shift
instruction with no users, terminating processing of the instruction.

Patch by: @brentdax (Brent Royal-Gordon)

Differential Revision: https://reviews.llvm.org/D66330

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369168 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/CodeGenPrepare.cpp b/lib/CodeGen/CodeGenPrepare.cpp
index d4d5fcd48cba94029f9e6e491552644dc488e055..7da810c40c48f987c4ad7c105c5c8b4e980ece3f 100644 (file)
--- a/lib/CodeGen/CodeGenPrepare.cpp
+++ b/lib/CodeGen/CodeGenPrepare.cpp
@@ -1682,10 +1682,11 @@ static bool OptimizeExtractBits(BinaryOperator *ShiftI, ConstantInt *CI,
     TheUse = InsertedShift;
   }
 
-  // If we removed all uses, nuke the shift.
+  // If we removed all uses, or there are none, nuke the shift.
   if (ShiftI->use_empty()) {
     salvageDebugInfo(*ShiftI);
     ShiftI->eraseFromParent();
+    MadeChange = true;
   }
 
   return MadeChange;

diff --git a/test/Transforms/CodeGenPrepare/sink-shift-and-trunc.ll b/test/Transforms/CodeGenPrepare/sink-shift-and-trunc.ll
index 07ea73b5a5ba765e4104b7f3dca63944a6334119..929a4f43e4abd09ad905dee01023e21873f94d97 100644 (file)
--- a/test/Transforms/CodeGenPrepare/sink-shift-and-trunc.ll
+++ b/test/Transforms/CodeGenPrepare/sink-shift-and-trunc.ll
@@ -58,6 +58,23 @@ return:                                           ; preds = %if.then17, %if.end1
   ret i32 %retval.0, !dbg !63
 }
 
+; CodeGenPrepare was erasing the unused lshr instruction, but then further
+; processing the instruction after it was freed. If this bug is still present,
+; this test will always crash in an LLVM built with ASAN enabled, and may
+; crash even if ASAN is not enabled.
+
+define i32 @shift_unused(i32 %a) {
+; CHECK-LABEL: @shift_unused(
+; CHECK-NEXT:  BB2:
+; CHECK-NEXT:    ret i32 [[A:%.*]]
+;
+  %as = lshr i32 %a, 3
+  br label %BB2
+
+BB2:
+  ret i32 %a
+}
+
 ; CHECK: [[shift1_loc]] = !DILocation(line: 1
 ; CHECK: [[trunc1_loc]] = !DILocation(line: 2
 ; CHECK: [[shift2_loc]] = !DILocation(line: 3