[PATCH] D49583: [x86/SLH] Rename and comment the main hardening function. NFC.

[Chandler Carruth via Phabricator via llvm-commits]

This revision was automatically updated to reflect the committed changes.
Closed by commit rL337667: [x86/SLH] Rename and comment the main hardening function. NFC. (authored by chandlerc, committed by ).

Changed prior to commit:
  https://reviews.llvm.org/D49583?vs=156412&id=156721#toc

Repository:
  rL LLVM

https://reviews.llvm.org/D49583

Files:
  llvm/trunk/lib/Target/X86/X86SpeculativeLoadHardening.cpp


Index: llvm/trunk/lib/Target/X86/X86SpeculativeLoadHardening.cpp
===================================================================
--- llvm/trunk/lib/Target/X86/X86SpeculativeLoadHardening.cpp
+++ llvm/trunk/lib/Target/X86/X86SpeculativeLoadHardening.cpp
@@ -168,7 +168,7 @@
   SmallVector<MachineInstr *, 16>
   tracePredStateThroughCFG(MachineFunction &MF, ArrayRef<BlockCondInfo> Infos);
 
-  void checkAllLoads(MachineFunction &MF);
+  void hardenAllLoads(MachineFunction &MF);
 
   unsigned saveEFLAGS(MachineBasicBlock &MBB,
                       MachineBasicBlock::iterator InsertPt, DebugLoc Loc);
@@ -504,8 +504,8 @@
     }
   }
 
-  // Now check all of the loads using the predicate state.
-  checkAllLoads(MF);
+  // Now harden all of the loads in the function using the predicate state.
+  hardenAllLoads(MF);
 
   // Now rewrite all the uses of the pred state using the SSA updater so that
   // we track updates through the CFG.
@@ -1224,7 +1224,24 @@
   return MBB.isLiveIn(X86::EFLAGS);
 }
 
-void X86SpeculativeLoadHardeningPass::checkAllLoads(MachineFunction &MF) {
+/// Harden all of the loads (including implicit loads) in the function.
+///
+/// This operates in two passes. First, we analyze the loads to determine which
+/// strategy will be used to harden them: hardening the address or hardening the
+/// loaded value when loaded into a register amenable to hardening. We have to
+/// process these first because the two strategies may interact -- later
+/// hardening may change what strategy we wish to use. We also will analyze
+/// data dependencies between loads and avoid hardening those loads that are
+/// data dependent on a load with a hardened address. We also skip hardening
+/// loads already behind an LFENCE as that is sufficient to harden them against
+/// misspeculation.
+///
+/// Second, we apply the hardening steps we determined necessary in the first
+/// pass.
+///
+/// These two passes are applied to each basic block. We operate one block at
+/// a time to simplify reasoning about reachability and sequencing.
+void X86SpeculativeLoadHardeningPass::hardenAllLoads(MachineFunction &MF) {
   // If the actual checking of loads is disabled, skip doing anything here.
   if (!HardenLoads)
     return;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D49583.156721.patch
Type: text/x-patch
Size: 2253 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180723/771e8a26/attachment-0001.bin>