Support -mnop-mcount and -mrecord-mcount on x86_64 elf targets. by pmur · Pull Request #184641 · llvm/llvm-project

pmur · 2026-03-04T16:38:06Z

I found out these attributes are only support on s390x when prototyping my Rust rfc for expanded instrumented function support.

Also, is this the correct/preferred way to insert nop's on x86?

Support mnop-mcount and mrecord-mcount attributes when used with the fentry-call=true attribute. Similar to SystemZ, use the mnop-mcount attribute to place a nop instead of a call to __fentry__. Likewise, use the mrecord-mcount attribute to record the location of the call to fentry (or nop if used with the above) to a section named __mcount_loc.

llvmbot · 2026-03-04T16:38:41Z

@llvm/pr-subscribers-clang-driver

@llvm/pr-subscribers-backend-x86

Author: Paul Murphy (pmur)

Changes

I found out these attributes are only support on s390x when prototyping my Rust rfc for expanded instrumented function support.

Also, is this the correct/preferred way to insert nop's on x86?

Full diff: https://github.com/llvm/llvm-project/pull/184641.diff

10 Files Affected:

(modified) clang/lib/Driver/ToolChains/Clang.cpp (+4-2)
(modified) clang/test/CodeGen/mnop-mcount.c (+8)
(modified) clang/test/CodeGen/mrecord-mcount.c (+8)
(modified) clang/test/Driver/mcount.c (+1-3)
(modified) llvm/lib/Target/X86/X86ISelDAGToDAG.cpp (+11-3)
(modified) llvm/lib/Target/X86/X86MCInstLower.cpp (+18)
(added) llvm/test/CodeGen/X86/mnop-mcount-01.ll (+26)
(added) llvm/test/CodeGen/X86/mnop-mcount-02.ll (+10)
(added) llvm/test/CodeGen/X86/mrecord-mcount-01.ll (+32)
(added) llvm/test/CodeGen/X86/mrecord-mcount-02.ll (+10)

diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index e860136aae5b3..2807be2c7edd0 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -6884,14 +6884,16 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
             << A->getAsString(Args) << TripleStr;
     }
     if (Arg *A = Args.getLastArg(options::OPT_mnop_mcount)) {
-      if (Arch == llvm::Triple::systemz)
+      if (Arch == llvm::Triple::systemz || (TC.getTriple().isX86_64() &&
+			                    TC.getTriple().isOSBinFormatELF()))
         A->render(Args, CmdArgs);
       else
         D.Diag(diag::err_drv_unsupported_opt_for_target)
             << A->getAsString(Args) << TripleStr;
     }
     if (Arg *A = Args.getLastArg(options::OPT_mrecord_mcount)) {
-      if (Arch == llvm::Triple::systemz)
+      if (Arch == llvm::Triple::systemz || (TC.getTriple().isX86_64() &&
+			                    TC.getTriple().isOSBinFormatELF()))
         A->render(Args, CmdArgs);
       else
         D.Diag(diag::err_drv_unsupported_opt_for_target)
diff --git a/clang/test/CodeGen/mnop-mcount.c b/clang/test/CodeGen/mnop-mcount.c
index ebb21cc560e37..31b4e995cff02 100644
--- a/clang/test/CodeGen/mnop-mcount.c
+++ b/clang/test/CodeGen/mnop-mcount.c
@@ -6,6 +6,14 @@
 // RUN:   %s 2>&1 | FileCheck -check-prefix=NOPG %s
 // RUN: %clang_cc1 -mnop-mcount -triple s390x-ibm-linux -emit-llvm -o - %s \
 // RUN:   2>&1 | FileCheck -check-prefix=NOPG %s
+// RUN: %clang_cc1 -pg -mfentry -mnop-mcount -triple x86_64-ibm-linux -emit-llvm \
+// RUN:   -o - %s 2>&1 | FileCheck  %s
+// RUN: not %clang_cc1 -pg -mnop-mcount -triple x86_64-ibm-linux -emit-llvm -o - \
+// RUN:   %s 2>&1 | FileCheck -check-prefix=NOMFENTRY %s
+// RUN: %clang_cc1 -mfentry -mnop-mcount -triple x86_64-ibm-linux -emit-llvm -o - \
+// RUN:   %s 2>&1 | FileCheck -check-prefix=NOPG %s
+// RUN: %clang_cc1 -mnop-mcount -triple x86_64-ibm-linux -emit-llvm -o - %s \
+// RUN:   2>&1 | FileCheck -check-prefix=NOPG %s
 
 int foo(void) {
   return 0;
diff --git a/clang/test/CodeGen/mrecord-mcount.c b/clang/test/CodeGen/mrecord-mcount.c
index 292a807073cef..832d4ee5e3e9f 100644
--- a/clang/test/CodeGen/mrecord-mcount.c
+++ b/clang/test/CodeGen/mrecord-mcount.c
@@ -6,6 +6,14 @@
 // RUN:   %s 2>&1 | FileCheck -check-prefix=NOPG %s
 // RUN: %clang_cc1 -mrecord-mcount -triple s390x-ibm-linux -emit-llvm -o - %s \
 // RUN:   2>&1 | FileCheck -check-prefix=NOPG %s
+// RUN: %clang_cc1 -pg -mfentry -mrecord-mcount -triple x86_64-ibm-linux -emit-llvm \
+// RUN:   -o - %s 2>&1 | FileCheck  %s
+// RUN: not %clang_cc1 -pg -mrecord-mcount -triple x86_64-ibm-linux -emit-llvm -o - \
+// RUN:   %s 2>&1 | FileCheck -check-prefix=NOMFENTRY %s
+// RUN: %clang_cc1 -mfentry -mrecord-mcount -triple x86_64-ibm-linux -emit-llvm -o - \
+// RUN:   %s 2>&1 | FileCheck -check-prefix=NOPG %s
+// RUN: %clang_cc1 -mrecord-mcount -triple x86_64-ibm-linux -emit-llvm -o - %s \
+// RUN:   2>&1 | FileCheck -check-prefix=NOPG %s
 
 int foo(void) {
   return 0;
diff --git a/clang/test/Driver/mcount.c b/clang/test/Driver/mcount.c
index 70f554745b9c3..72d48b3806fe7 100644
--- a/clang/test/Driver/mcount.c
+++ b/clang/test/Driver/mcount.c
@@ -1,12 +1,10 @@
 // RUN: %clang --target=s390x -c -### %s -mnop-mcount -mrecord-mcount 2>&1 | FileCheck %s
+// RUN: %clang --target=x86_64 -c -### %s -mnop-mcount -mrecord-mcount 2>&1 | FileCheck %s
 
 // CHECK: "-mnop-mcount"
 // CHECK: "-mrecord-mcount"
 
-// RUN: not %clang --target=x86_64 -c -### %s -mnop-mcount -mrecord-mcount 2>&1 | FileCheck --check-prefix=ERR1 %s
 // RUN: not %clang --target=aarch64 -c -### %s -mnop-mcount -mrecord-mcount 2>&1 | FileCheck --check-prefix=ERR2 %s
 
-// ERR1: error: unsupported option '-mnop-mcount' for target 'x86_64'
-// ERR1: error: unsupported option '-mrecord-mcount' for target 'x86_64'
 // ERR2: error: unsupported option '-mnop-mcount' for target 'aarch64'
 // ERR2: error: unsupported option '-mrecord-mcount' for target 'aarch64'
diff --git a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
index 101ea3e231a5c..aa7efb8a05e1c 100644
--- a/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
+++ b/llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
@@ -179,11 +179,19 @@ namespace {
     bool runOnMachineFunction(MachineFunction &MF) override {
       // Reset the subtarget each time through.
       Subtarget = &MF.getSubtarget<X86Subtarget>();
-      IndirectTlsSegRefs = MF.getFunction().hasFnAttribute(
-                             "indirect-tls-seg-refs");
+      const Function &F = MF.getFunction();
+
+      IndirectTlsSegRefs = F.hasFnAttribute("indirect-tls-seg-refs");
+
+      if (F.getFnAttribute("fentry-call").getValueAsString() != "true") {
+        if (F.hasFnAttribute("mnop-mcount"))
+          report_fatal_error("mnop-mcount only supported with fentry-call");
+        if (F.hasFnAttribute("mrecord-mcount"))
+          report_fatal_error("mrecord-mcount only supported with fentry-call");
+      }
 
       // OptFor[Min]Size are used in pattern predicates that isel is matching.
-      OptForMinSize = MF.getFunction().hasMinSize();
+      OptForMinSize = F.hasMinSize();
       return SelectionDAGISel::runOnMachineFunction(MF);
     }
 
diff --git a/llvm/lib/Target/X86/X86MCInstLower.cpp b/llvm/lib/Target/X86/X86MCInstLower.cpp
index 4ef4718e1b01c..8bbd40d973fab 100644
--- a/llvm/lib/Target/X86/X86MCInstLower.cpp
+++ b/llvm/lib/Target/X86/X86MCInstLower.cpp
@@ -26,6 +26,7 @@
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/StringExtras.h"
+#include "llvm/BinaryFormat/ELF.h"
 #include "llvm/CodeGen/MachineBranchProbabilityInfo.h"
 #include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -44,6 +45,7 @@
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstBuilder.h"
 #include "llvm/MC/MCSection.h"
+#include "llvm/MC/MCSectionELF.h"
 #include "llvm/MC/MCStreamer.h"
 #include "llvm/MC/MCSymbol.h"
 #include "llvm/MC/TargetRegistry.h"
@@ -885,6 +887,22 @@ void X86AsmPrinter::LowerFENTRY_CALL(const MachineInstr &MI,
                                      X86MCInstLower &MCIL) {
   bool Is64Bits = Subtarget->is64Bit();
   MCContext &Ctx = OutStreamer->getContext();
+
+  if (MF->getFunction().hasFnAttribute("mrecord-mcount")) {
+    MCSymbol *DotSym = OutContext.createTempSymbol();
+    OutStreamer->pushSection();
+    OutStreamer->switchSection(
+        Ctx.getELFSection("__mcount_loc", ELF::SHT_PROGBITS, ELF::SHF_ALLOC));
+    OutStreamer->emitSymbolValue(DotSym, Is64Bits ? 8 : 4);
+    OutStreamer->popSection();
+    OutStreamer->emitLabel(DotSym);
+  }
+
+  if (MF->getFunction().hasFnAttribute("mnop-mcount")) {
+    emitX86Nops(*OutStreamer, 5, Subtarget);
+    return;
+  }
+
   MCSymbol *fentry = Ctx.getOrCreateSymbol("__fentry__");
   const MCSymbolRefExpr *Op = MCSymbolRefExpr::create(fentry, Ctx);
 
diff --git a/llvm/test/CodeGen/X86/mnop-mcount-01.ll b/llvm/test/CodeGen/X86/mnop-mcount-01.ll
new file mode 100644
index 0000000000000..ddefa497659f3
--- /dev/null
+++ b/llvm/test/CodeGen/X86/mnop-mcount-01.ll
@@ -0,0 +1,26 @@
+; RUN: llc %s -mtriple=x86_64-linux-gnu -o - -verify-machineinstrs \
+; RUN:   | FileCheck %s
+
+define void @test1() #0 {
+entry:
+  ret void
+
+; CHECK-LABEL: @test1
+; CHECK: callq __fentry__
+; CHECK-NOT: nopl 8(%rax,%rax)
+; CHECK: retq
+}
+
+define void @test2() #1 {
+entry:
+  ret void
+
+; CHECK-LABEL: @test2
+; CHECK-NOT: callq __fentry__
+; CHECK: nopl 8(%rax,%rax)
+; CHECK: retq
+}
+
+attributes #0 = { "fentry-call"="true" }
+attributes #1 = { "fentry-call"="true" "mnop-mcount" }
+
diff --git a/llvm/test/CodeGen/X86/mnop-mcount-02.ll b/llvm/test/CodeGen/X86/mnop-mcount-02.ll
new file mode 100644
index 0000000000000..d576e0f289818
--- /dev/null
+++ b/llvm/test/CodeGen/X86/mnop-mcount-02.ll
@@ -0,0 +1,10 @@
+; RUN: not --crash llc %s -mtriple=x86_64-linux-gnu -o - 2>&1 | FileCheck %s
+;
+; CHECK: LLVM ERROR: mnop-mcount only supported with fentry-call
+
+define void @test1() #0 {
+entry:
+  ret void
+}
+
+attributes #0 = { "instrument-function-entry-inlined"="mcount" "mnop-mcount" }
diff --git a/llvm/test/CodeGen/X86/mrecord-mcount-01.ll b/llvm/test/CodeGen/X86/mrecord-mcount-01.ll
new file mode 100644
index 0000000000000..6202a652b8831
--- /dev/null
+++ b/llvm/test/CodeGen/X86/mrecord-mcount-01.ll
@@ -0,0 +1,32 @@
+; RUN: llc %s -mtriple=x86_64-linux-gnu -o - -verify-machineinstrs \
+; RUN:   | FileCheck %s
+
+define void @test1() #0 {
+entry:
+  ret void
+
+; CHECK-LABEL: test1:
+; CHECK: .section __mcount_loc,"a",@progbits
+; CHECK: .quad .Ltmp0
+; CHECK: .text
+; CHECK: .Ltmp0:
+; CHECK: callq __fentry__
+; CHECK: retq 
+}
+
+define void @test2() #1 {
+entry:
+  ret void
+
+; CHECK-LABEL: test2:
+; CHECK: .section __mcount_loc,"a",@progbits
+; CHECK: .quad .Ltmp1
+; CHECK: .text
+; CHECK: .Ltmp1:
+; CHECK: nopl 8(%rax,%rax)
+; CHECK-NOT: callq __fentry__
+; CHECK: retq
+}
+
+attributes #0 = { "fentry-call"="true" "mrecord-mcount" }
+attributes #1 = { "fentry-call"="true" "mnop-mcount" "mrecord-mcount" }
diff --git a/llvm/test/CodeGen/X86/mrecord-mcount-02.ll b/llvm/test/CodeGen/X86/mrecord-mcount-02.ll
new file mode 100644
index 0000000000000..a1f80f1840c41
--- /dev/null
+++ b/llvm/test/CodeGen/X86/mrecord-mcount-02.ll
@@ -0,0 +1,10 @@
+; RUN: not --crash llc %s -mtriple=x86_64-linux-gnu -o - 2>&1 | FileCheck %s
+;
+; CHECK: LLVM ERROR: mrecord-mcount only supported with fentry-call
+
+define void @test1() #0 {
+entry:
+  ret void
+}
+
+attributes #0 = { "instrument-function-entry-inlined"="mcount" "mrecord-mcount" }

github-actions · 2026-03-04T16:40:01Z

✅ With the latest revision this PR passed the C/C++ code formatter.

These are gated by -mfentry.

nikic

How does this differ from the "patchable-function-entry" attribute? I believe that one has good cross target support for inserting function entry nops on at least X86, AArch64, PowerPC, RISCV, LoongArch and SystemZ.

nikic · 2026-03-26T16:20:17Z

Added some X86 reviewers, but I think this could use more context on why we want to support these options, and why we want to support them on s390x and x86 in particular. Is it just because those are the two arches where GCC supports them?

pmur · 2026-03-27T14:32:44Z

How does this differ from the "patchable-function-entry" attribute? I believe that one has good cross target support for inserting function entry nops on at least X86, AArch64, PowerPC, RISCV, LoongArch and SystemZ.

I suppose fentry is a more specialized variant. Interestingly, the x86-64 kernel uses both. It seems to be a crude way to separate bpf and ftrace usage of patchable entries.

Of minor note, the gcc x86-64 windows targets default to using fentry over mcount when using -pg.

llvmbot added backend:X86 clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' labels Mar 4, 2026

[clang] enable -mnop-mcount -mrecord-mcount for x86_64

27616c7

These are gated by -mfentry.

pmur force-pushed the murp/support-mcount-attr-x86 branch from 0c14b0a to 27616c7 Compare March 4, 2026 20:41

nikic reviewed Mar 26, 2026

View reviewed changes

nikic requested review from RKSimon and phoebewang March 26, 2026 16:19

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support -mnop-mcount and -mrecord-mcount on x86_64 elf targets.#184641

Support -mnop-mcount and -mrecord-mcount on x86_64 elf targets.#184641
pmur wants to merge 2 commits intollvm:mainfrom
pmur:murp/support-mcount-attr-x86

pmur commented Mar 4, 2026

Uh oh!

llvmbot commented Mar 4, 2026 •

edited

Loading

Uh oh!

github-actions bot commented Mar 4, 2026 •

edited

Loading

Uh oh!

nikic left a comment

Uh oh!

nikic commented Mar 26, 2026

Uh oh!

pmur commented Mar 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

pmur commented Mar 4, 2026

Uh oh!

llvmbot commented Mar 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Mar 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nikic left a comment

Choose a reason for hiding this comment

Uh oh!

nikic commented Mar 26, 2026

Uh oh!

pmur commented Mar 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

llvmbot commented Mar 4, 2026 •

edited

Loading

github-actions bot commented Mar 4, 2026 •

edited

Loading