<html>
<head>
<base href="https://llvm.org/bugs/" />
</head>
<body><table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Bug ID</th>
<td><a class="bz_bug_link
bz_status_NEW "
title="NEW --- - expandMOVImm for AArch64 can in rare cases assign an immediate to SP instead the desired WZR pseudo register"
href="https://llvm.org/bugs/show_bug.cgi?id=23209">23209</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>expandMOVImm for AArch64 can in rare cases assign an immediate to SP instead the desired WZR pseudo register
</td>
</tr>
<tr>
<th>Product</th>
<td>libraries
</td>
</tr>
<tr>
<th>Version</th>
<td>3.6
</td>
</tr>
<tr>
<th>Hardware</th>
<td>Macintosh
</td>
</tr>
<tr>
<th>OS</th>
<td>MacOS X
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>normal
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>Backend: AArch64
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>por@cryptomathic.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvmbugs@cs.uiuc.edu
</td>
</tr>
<tr>
<th>Classification</th>
<td>Unclassified
</td>
</tr></table>
<p>
<div>
<pre>Hi,
I have seen a few cases of expandMOVImm on AArch64 generating an ORRWri
instruction writing to SP where WZR was intended, causing crashes at run-time.
The problem has been observed with LLVM 3.5.1 and LLVM 3.6.0.
The following patch seems to fix it for us:
-------------------------
Index: AArch64ExpandPseudoInsts.cpp
===================================================================
--- AArch64ExpandPseudoInsts.cpp (revision 125567)
+++ AArch64ExpandPseudoInsts.cpp (revision 125568)
@@ -398,6 +398,31 @@
// Try a MOVI instruction (aka ORR-immediate with the zero register).
uint64_t UImm = Imm << (64 - BitSize) >> (64 - BitSize);
uint64_t Encoding;
+ const unsigned DstReg0 = MI.getOperand(0).getReg();
+
+ // explicit tests for MOV ZR,imm which *can* be generated by morpher and
must be removed,
+ // it's logically a no-op, but since ZR (x31) is an alias for SP (x31) in
many instructions like ORR
+ // it will clobber the SP...
+ if (BitSize == 32) {
+ // moving an immediate to WZR is a logical no-op, remove it
+ if (DstReg0 == AArch64::WZR) {
+ MI.eraseFromParent();
+ return true;
+ }
+ assert(DstReg0 != AArch64::WSP && "Mov imm to WSP forbidden");
+ assert(DstReg0 != AArch64::WZR && "Mov imm to WZR forbidden");
+ assert(AArch64::GPR32spRegClass.contains(DstReg0) && "Mov dest not 32
bit SP reg class");
+ } else {
+ // moving an immediate to XZR is a logical no-op, remove it
+ if (DstReg0 == AArch64::XZR) {
+ MI.eraseFromParent();
+ return true;
+ }
+ assert(DstReg0 != AArch64::SP && "Mov imm to SP forbidden");
+ assert(DstReg0 != AArch64::XZR && "Mov imm to XZR forbidden");
+ assert(AArch64::GPR64spRegClass.contains(DstReg0) && "Mov dest not 64
bit SP reg class");
+ }
+
if (AArch64_AM::processLogicalImmediate(UImm, BitSize, Encoding)) {
unsigned Opc = (BitSize == 32 ? AArch64::ORRWri : AArch64::ORRXri);
MachineInstrBuilder MIB =
----------------------------</pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are on the CC list for the bug.</li>
</ul>
</body>
</html>