[llvm] [BOLT] Drop -znow requirement for PLT optimization on x86-64 (PR #178758)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-bolt

Author: Maksim Panchenko (maksfb)

<details>
<summary>Changes</summary>

On x86-64, PLT optimization does not require the binary to be linked with -znow because indirect calls through GOT work correctly with lazy binding. At runtime, the dynamic linker's resolver will populate the GOT entry on the first call, just like with a regular PLT call.

This change removes the -znow requirement specifically for x86-64 while keeping it for other architectures. I haven't checked RISV-V, but it's still necessary on AArch64.

---
Full diff: https://github.com/llvm/llvm-project/pull/178758.diff


3 Files Affected:

- (modified) bolt/docs/CommandLineArgumentReference.md (+1-1) 
- (modified) bolt/lib/Passes/PLTCall.cpp (+9-2) 
- (added) bolt/test/runtime/X86/plt-call.c (+14) 


``````````diff
diff --git a/bolt/docs/CommandLineArgumentReference.md b/bolt/docs/CommandLineArgumentReference.md
index 0dbf6f59d5e88..bf706ba3a1cdc 100644
--- a/bolt/docs/CommandLineArgumentReference.md
+++ b/bolt/docs/CommandLineArgumentReference.md
@@ -724,7 +724,7 @@
 
 - `--plt=<value>`
 
-  Optimize PLT calls (requires linking with -znow)
+  Optimize PLT calls (requires linking with -znow on non-x86 architectures)
   - `none`: do not optimize PLT calls
   - `hot`: optimize executed (hot) PLT calls
   - `all`: optimize all PLT calls
diff --git a/bolt/lib/Passes/PLTCall.cpp b/bolt/lib/Passes/PLTCall.cpp
index 90b5f586a7bad..a41b3870af666 100644
--- a/bolt/lib/Passes/PLTCall.cpp
+++ b/bolt/lib/Passes/PLTCall.cpp
@@ -23,7 +23,9 @@ namespace opts {
 extern cl::OptionCategory BoltOptCategory;
 
 static cl::opt<bolt::PLTCall::OptType>
-    PLT("plt", cl::desc("optimize PLT calls (requires linking with -znow)"),
+    PLT("plt",
+        cl::desc("optimize PLT calls (requires linking with -znow on "
+                 "non-x86 architectures)"),
         cl::init(bolt::PLTCall::OT_NONE),
         cl::values(clEnumValN(bolt::PLTCall::OT_NONE, "none",
                               "do not optimize PLT calls"),
@@ -76,7 +78,12 @@ Error PLTCall::runOnFunctions(BinaryContext &BC) {
   }
 
   if (NumCallsOptimized) {
-    BC.RequiresZNow = true;
+    // On X86-64, PLT optimization does not require -znow because the indirect
+    // call through GOT works correctly with lazy binding. At runtime, the
+    // resolver will populate the GOT entry on first call just like with a
+    // regular PLT call.
+    if (!BC.isX86())
+      BC.RequiresZNow = true;
     BC.outs() << "BOLT-INFO: " << NumCallsOptimized
               << " PLT calls in the binary were optimized.\n";
   }
diff --git a/bolt/test/runtime/X86/plt-call.c b/bolt/test/runtime/X86/plt-call.c
new file mode 100644
index 0000000000000..3ece4c7c16cca
--- /dev/null
+++ b/bolt/test/runtime/X86/plt-call.c
@@ -0,0 +1,14 @@
+// Check that PLT optimization works on X86-64 without -znow.
+
+// RUN: %clang %cflags %s -o %t.exe -Wl,-q
+// RUN: llvm-bolt %t.exe -o %t.bolt --plt=all
+// RUN: %t.bolt | FileCheck %s
+
+// CHECK: Success
+
+#include <stdio.h>
+
+int main() {
+  puts("Success");
+  return 0;
+}

``````````

</details>


https://github.com/llvm/llvm-project/pull/178758