[X86] Disable speculative load hardening for operations with an explicit RSP base.
After D58632, we can create idempotent atomic operations to the top of stack.
This confused speculative load hardening because it thinks accesses should have
virtual register base except for the cases it already excluded.
This commit adds a new exclusion for this case. I'll try to reduce a test case
for this, but this fix was verified to work by the reporter. This should avoid
needing to revert D58632.
llvm-svn: 360475
diff --git a/llvm/lib/Target/X86/X86SpeculativeLoadHardening.cpp b/llvm/lib/Target/X86/X86SpeculativeLoadHardening.cpp
index c8b740c..af0c763 100644
--- a/llvm/lib/Target/X86/X86SpeculativeLoadHardening.cpp
+++ b/llvm/lib/Target/X86/X86SpeculativeLoadHardening.cpp
@@ -1964,6 +1964,14 @@
LLVM_DEBUG(
dbgs() << " Skipping hardening base of explicit stack frame load: ";
MI.dump(); dbgs() << "\n");
+ } else if (BaseMO.getReg() == X86::RSP) {
+ // Some idempotent atomic operations are lowered directly to a locked
+ // OR with 0 to the top of stack(or slightly offset from top) which uses an
+ // explicit RSP register as the base.
+ assert(IndexMO.getReg() == X86::NoRegister &&
+ "Explicit RSP access with dynamic index!");
+ LLVM_DEBUG(
+ dbgs() << " Cannot harden base of explicit RSP offset in a load!");
} else if (BaseMO.getReg() == X86::RIP ||
BaseMO.getReg() == X86::NoRegister) {
// For both RIP-relative addressed loads or absolute loads, we cannot