[BOLT] Fix trap value for non-X86

treapster · treapster · commit 28fd2ca14246 · 2023-08-24T01:29:41.000+03:00
The trap value used by BOLT was assumed to be single-byte instruction. It made some functions unaligned on AArch64(e.g exceptions-instrumentation test) and caused emission failures. Fix that by changing fill value to StringRef. Reviewed By: rafauler Differential Revision: https://reviews.llvm.org/D158191
diff --git a/bolt/include/bolt/Core/MCPlusBuilder.h b/bolt/include/bolt/Core/MCPlusBuilder.h
@@ -639,9 +639,12 @@ class MCPlusBuilder {
     return false;
   }
 
-  /// If non-zero, this is used to fill the executable space with instructions
-  /// that will trap. Defaults to 0.
-  virtual unsigned getTrapFillValue() const { return 0; }
+  /// Used to fill the executable space with instructions
+  /// that will trap.
+  virtual StringRef getTrapFillValue() const {
+    llvm_unreachable("not implemented");
+    return StringRef();
+  }
 
   /// Interface and basic functionality of a MCInstMatcher. The idea is to make
   /// it easy to match one or more MCInsts against a tree-like pattern and
diff --git a/bolt/lib/Core/BinaryEmitter.cpp b/bolt/lib/Core/BinaryEmitter.cpp
@@ -380,7 +380,7 @@ bool BinaryEmitter::emitFunction(BinaryFunction &Function,
   }
 
   if (opts::MarkFuncs)
-    Streamer.emitIntValue(BC.MIB->getTrapFillValue(), 1);
+    Streamer.emitBytes(BC.MIB->getTrapFillValue());
 
   // Emit CFI end
   if (Function.hasCFI())
@@ -424,7 +424,7 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,
     // case, the call site entries in that LSDA have 0 as offset to the landing
     // pad, which the runtime interprets as "no handler". To prevent this,
     // insert some padding.
-    Streamer.emitIntValue(BC.MIB->getTrapFillValue(), 1);
+    Streamer.emitBytes(BC.MIB->getTrapFillValue());
   }
 
   // Track the first emitted instruction with debug info.
diff --git a/bolt/lib/Rewrite/RewriteInstance.cpp b/bolt/lib/Rewrite/RewriteInstance.cpp
@@ -5286,8 +5286,10 @@ void RewriteInstance::rewriteFile() {
       if (!BF.getFileOffset() || !BF.isEmitted())
         continue;
       OS.seek(BF.getFileOffset());
-      for (unsigned I = 0; I < BF.getMaxSize(); ++I)
-        OS.write((unsigned char)BC->MIB->getTrapFillValue());
+      StringRef TrapInstr = BC->MIB->getTrapFillValue();
+      unsigned NInstr = BF.getMaxSize() / TrapInstr.size();
+      for (unsigned I = 0; I < NInstr; ++I)
+        OS.write(TrapInstr.data(), TrapInstr.size());
     }
     OS.seek(SavedPos);
   }
diff --git a/bolt/lib/Target/AArch64/AArch64MCPlusBuilder.cpp b/bolt/lib/Target/AArch64/AArch64MCPlusBuilder.cpp
@@ -1134,6 +1134,10 @@ class AArch64MCPlusBuilder : public MCPlusBuilder {
     }
   }
 
+  StringRef getTrapFillValue() const override {
+    return StringRef("\0\0\0\0", 4);
+  }
+
   bool createReturn(MCInst &Inst) const override {
     Inst.setOpcode(AArch64::RET);
     Inst.clear();
diff --git a/bolt/lib/Target/RISCV/RISCVMCPlusBuilder.cpp b/bolt/lib/Target/RISCV/RISCVMCPlusBuilder.cpp
@@ -171,6 +171,10 @@ class RISCVMCPlusBuilder : public MCPlusBuilder {
     return true;
   }
 
+  StringRef getTrapFillValue() const override {
+    return StringRef("\0\0\0\0", 4);
+  }
+
   bool analyzeBranch(InstructionIterator Begin, InstructionIterator End,
                      const MCSymbol *&TBB, const MCSymbol *&FBB,
                      MCInst *&CondBranch,
diff --git a/bolt/lib/Target/X86/X86MCPlusBuilder.cpp b/bolt/lib/Target/X86/X86MCPlusBuilder.cpp
@@ -397,7 +397,7 @@ class X86MCPlusBuilder : public MCPlusBuilder {
     }
   }
 
-  unsigned getTrapFillValue() const override { return 0xCC; }
+  StringRef getTrapFillValue() const override { return StringRef("\314", 1); }
 
   struct IndJmpMatcherFrag1 : MCInstMatcher {
     std::unique_ptr<MCInstMatcher> Base;

Original file line number	Diff line number	Diff line change
`@@ -380,7 +380,7 @@ bool BinaryEmitter::emitFunction(BinaryFunction &Function,`
`380`	`380`	`}`
`381`	`381`
`382`	`382`	`if (opts::MarkFuncs)`
`383`		`- Streamer.emitIntValue(BC.MIB->getTrapFillValue(), 1);`
	`383`	`+ Streamer.emitBytes(BC.MIB->getTrapFillValue());`
`384`	`384`
`385`	`385`	`// Emit CFI end`
`386`	`386`	`if (Function.hasCFI())`
`@@ -424,7 +424,7 @@ void BinaryEmitter::emitFunctionBody(BinaryFunction &BF, FunctionFragment &FF,`
`424`	`424`	`// case, the call site entries in that LSDA have 0 as offset to the landing`
`425`	`425`	`// pad, which the runtime interprets as "no handler". To prevent this,`
`426`	`426`	`// insert some padding.`
`427`		`- Streamer.emitIntValue(BC.MIB->getTrapFillValue(), 1);`
	`427`	`+ Streamer.emitBytes(BC.MIB->getTrapFillValue());`
`428`	`428`	`}`
`429`	`429`
`430`	`430`	`// Track the first emitted instruction with debug info.`
Original file line number	Diff line number	Diff line change
`@@ -5286,8 +5286,10 @@ void RewriteInstance::rewriteFile() {`
`5286`	`5286`	`if (!BF.getFileOffset() \|\| !BF.isEmitted())`
`5287`	`5287`	`continue;`
`5288`	`5288`	`OS.seek(BF.getFileOffset());`
`5289`		`- for (unsigned I = 0; I < BF.getMaxSize(); ++I)`
`5290`		`- OS.write((unsigned char)BC->MIB->getTrapFillValue());`
	`5289`	`+ StringRef TrapInstr = BC->MIB->getTrapFillValue();`
	`5290`	`+ unsigned NInstr = BF.getMaxSize() / TrapInstr.size();`
	`5291`	`+ for (unsigned I = 0; I < NInstr; ++I)`
	`5292`	`+ OS.write(TrapInstr.data(), TrapInstr.size());`
`5291`	`5293`	`}`
`5292`	`5294`	`OS.seek(SavedPos);`
`5293`	`5295`	`}`
Original file line number	Diff line number	Diff line change
`@@ -1134,6 +1134,10 @@ class AArch64MCPlusBuilder : public MCPlusBuilder {`
`1134`	`1134`	`}`
`1135`	`1135`	`}`
`1136`	`1136`
	`1137`	`+ StringRef getTrapFillValue() const override {`
	`1138`	`+ return StringRef("\0\0\0\0", 4);`
	`1139`	`+ }`
	`1140`	`+`
`1137`	`1141`	`bool createReturn(MCInst &Inst) const override {`
`1138`	`1142`	`Inst.setOpcode(AArch64::RET);`
`1139`	`1143`	`Inst.clear();`
Original file line number	Diff line number	Diff line change
`@@ -397,7 +397,7 @@ class X86MCPlusBuilder : public MCPlusBuilder {`
`397`	`397`	`}`
`398`	`398`	`}`
`399`	`399`
`400`		`- unsigned getTrapFillValue() const override { return 0xCC; }`
	`400`	`+ StringRef getTrapFillValue() const override { return StringRef("\314", 1); }`
`401`	`401`
`402`	`402`	`struct IndJmpMatcherFrag1 : MCInstMatcher {`
`403`	`403`	`std::unique_ptr<MCInstMatcher> Base;`