[llvm-branch-commits] [llvm] [BOLT] Gadget scanner: Detect address materialization and arithmetics (PR #132540)

[Jacob Bramley via llvm-branch-commits]

================
@@ -0,0 +1,228 @@
+// RUN: %clang %cflags -march=armv8.3-a %s -o %t.exe
+// RUN: llvm-bolt-binary-analysis --scanners=pauth %t.exe 2>&1 | FileCheck %s
+
+// Test various patterns that should or should not be considered safe
+// materialization of PC-relative addresses.
+//
+// Note that while "instructions that write to the affected registers"
+// section of the report is still technically correct, it does not necessarily
+// mentions the instructions that are used incorrectly.
+//
+// FIXME: Switch to PAC* instructions instead of indirect tail call for testing
+//        if a register is considered safe when detection of signing oracles is
+//        implemented, as it is more traditional usage of PC-relative constants.
+//        Moreover, using PAC instructions would improve test robustness, as
+//        handling of *calls* can be influenced by what BOLT classifies as a
+//        tail call, for example.
+
+        .text
+
----------------
jacobbramley wrote:

I don't think there are any tests for `sub`. I doubt that it's likely to be much different from `add`, but it's probably worth having one case.

https://github.com/llvm/llvm-project/pull/132540