[SystemZ] Add proper mcount handling #135767

dominik-steenken · 2025-04-15T09:42:30Z

When compiling with -pg, the EntryExitInstrumenterPass will insert calls to the glibc function mcount at the begining of each MachineFunction.

On SystemZ, these calls require special handling:

The call to mcount needs to happen at the beginning of the prologue.
Prior to the call to mcount, register %r14, the return address of the callee function, must be stored 8 bytes above the stack pointer %r15. After the call to mcount returns, that register needs to be restored.

This commit adds some special handling to the EntryExitInstrumenterPass that keeps the insertion of the mcount function into the module, but skips over insertion of the actual call in order to perform this insertion in the emitPrologue function. There, a simple sequence of store/call/load is inserted, which implements the above.

The desired change in the EntryExitInstrumenterPass necessitated the addition of a new attribute and attribute kind to each function, which is used to trigger the postprocessing, aka call insertion, in emitPrologue. Note that the new attribute must be of a different kind than the mcount atribute, since otherwise it would replace that attribute and later be deleted by the code that intended to delete mcount. The new attribnute is called insert-mcount, while the attribute kind is systemz-backend, to clearly mark it as a SystemZ-specific backend concern.

This PR should address issue #121137 . The test inserted here is derived from the example given in that issue.

llvmbot · 2025-04-15T09:43:06Z

@llvm/pr-subscribers-backend-systemz

Author: Dominik Steenken (dominik-steenken)

Changes

When compiling with -pg, the EntryExitInstrumenterPass will insert calls to the glibc function mcount at the begining of each MachineFunction.

On SystemZ, these calls require special handling:

The call to mcount needs to happen at the beginning of the prologue.
Prior to the call to mcount, register %r14, the return address of the callee function, must be stored 8 bytes above the stack pointer %r15. After the call to mcount returns, that register needs to be restored.

This commit adds some special handling to the EntryExitInstrumenterPass that keeps the insertion of the mcount function into the module, but skips over insertion of the actual call in order to perform this insertion in the emitPrologue function. There, a simple sequence of store/call/load is inserted, which implements the above.

Noe that the desired change in the EntryExitInstrumenterPass necessitated a change to the signature of the insertCall function, since there now is a possibility that no call is inserted, and thus that needs to be communicated to the caller. We can't ignore that fact either, because we need Changed to be false in the caller for SystemZ, so we can use it in a guard to prevent the deletion of the mcount attribute from the function, which we still need for our postprocessing, aka call insertion, in emitPrologue.

This PR should address issue #121137 . The test inserted here is derived from the example given in that issue.

Full diff: https://github.com/llvm/llvm-project/pull/135767.diff

3 Files Affected:

(modified) llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp (+28)
(modified) llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp (+10-6)
(added) llvm/test/CodeGen/SystemZ/mcount.ll (+35)

diff --git a/llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp b/llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp
index 9561ea544b270..0936690dafdcd 100644
--- a/llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp
+++ b/llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp
@@ -7,6 +7,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "SystemZFrameLowering.h"
+#include "MCTargetDesc/SystemZMCTargetDesc.h"
 #include "SystemZCallingConv.h"
 #include "SystemZInstrInfo.h"
 #include "SystemZMachineFunctionInfo.h"
@@ -18,6 +19,8 @@
 #include "llvm/CodeGen/RegisterScavenging.h"
 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
 #include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Target/TargetMachine.h"
 
 using namespace llvm;
@@ -558,6 +561,31 @@ void SystemZELFFrameLowering::emitPrologue(MachineFunction &MF,
   // to determine the end of the prologue.
   DebugLoc DL;
 
+  // Add mcount instrumentation if necessary.
+  if (MF.getFunction().getFnAttribute("instrument-function-entry-inlined").getValueAsString() == "mcount") {
+
+    // Store return address 8 bytes above stack pointer.
+    BuildMI(MBB, MBBI, DL, ZII->get(SystemZ::STG))
+        .addReg(SystemZ::R14D)
+        .addReg(SystemZ::R15D)
+        .addImm(8)
+        .addReg(0);
+
+    // Call mcount (Regmask 0 to ensure this will not be moved by the
+    // scheduler.).
+    const uint32_t Mask = 0;
+    BuildMI(MBB, MBBI, DL, ZII->get(SystemZ::CallBRASL))
+        .addGlobalAddress(MF.getFunction().getParent()->getFunction("mcount"))
+        .addRegMask(&Mask);
+
+    // Reload return address drom 8 bytes above stack pointer.
+    BuildMI(MBB, MBBI, DL, ZII->get(SystemZ::LG))
+      .addReg(SystemZ::R14D)
+      .addReg(SystemZ::R15D)
+      .addImm(8)
+      .addReg(0);
+  }
+
   // The current offset of the stack pointer from the CFA.
   int64_t SPOffsetFromCFA = -SystemZMC::ELFCFAOffsetFromInitialSP;
 
diff --git a/llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp b/llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp
index d47f1b4253b54..72e4f8791a91a 100644
--- a/llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp
+++ b/llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp
@@ -22,7 +22,7 @@
 
 using namespace llvm;
 
-static void insertCall(Function &CurFn, StringRef Func,
+static bool insertCall(Function &CurFn, StringRef Func,
                        BasicBlock::iterator InsertionPt, DebugLoc DL) {
   Module &M = *InsertionPt->getParent()->getParent()->getParent();
   LLVMContext &C = InsertionPt->getParent()->getContext();
@@ -63,12 +63,16 @@ static void insertCall(Function &CurFn, StringRef Func,
                                   false));
       CallInst *Call = CallInst::Create(Fn, RetAddr, "", InsertionPt);
       Call->setDebugLoc(DL);
+    } else if (TargetTriple.isSystemZ()) {
+      M.getOrInsertFunction(Func, Type::getVoidTy(C));
+      // skip insertion for `mcount` on SystemZ. This will be handled later in `emitPrologue`.
+      return false;
     } else {
       FunctionCallee Fn = M.getOrInsertFunction(Func, Type::getVoidTy(C));
       CallInst *Call = CallInst::Create(Fn, "", InsertionPt);
       Call->setDebugLoc(DL);
     }
-    return;
+    return true;
   }
 
   if (Func == "__cyg_profile_func_enter" || Func == "__cyg_profile_func_exit") {
@@ -87,7 +91,7 @@ static void insertCall(Function &CurFn, StringRef Func,
     CallInst *Call =
         CallInst::Create(Fn, ArrayRef<Value *>(Args), "", InsertionPt);
     Call->setDebugLoc(DL);
-    return;
+    return true;
   }
 
   // We only know how to call a fixed set of instrumentation functions, because
@@ -129,9 +133,9 @@ static bool runOnFunction(Function &F, bool PostInlining) {
     if (auto SP = F.getSubprogram())
       DL = DILocation::get(SP->getContext(), SP->getScopeLine(), 0, SP);
 
-    insertCall(F, EntryFunc, F.begin()->getFirstInsertionPt(), DL);
-    Changed = true;
-    F.removeFnAttr(EntryAttr);
+    Changed = insertCall(F, EntryFunc, F.begin()->getFirstInsertionPt(), DL);
+    if (Changed)
+      F.removeFnAttr(EntryAttr);
   }
 
   if (!ExitFunc.empty()) {
diff --git a/llvm/test/CodeGen/SystemZ/mcount.ll b/llvm/test/CodeGen/SystemZ/mcount.ll
new file mode 100644
index 0000000000000..01bd34548f125
--- /dev/null
+++ b/llvm/test/CodeGen/SystemZ/mcount.ll
@@ -0,0 +1,35 @@
+; Test proper insertion of mcount instrumentation
+;
+; RUN: llc < %s -mtriple=s390x-linux-gnu -o - | FileCheck %s
+;
+; CHECK: # %bb.0:
+; CHECK-NEXT: stg %r14, 8(%r15)
+; CHECK-NEXT: brasl %r14, mcount@PLT
+; CHECK-NEXT: lg %r14, 8(%r15)
+define dso_local signext i32 @fib(i32 noundef signext %n) #0 {
+entry:
+  %n.addr = alloca i32, align 4
+  store i32 %n, ptr %n.addr, align 4
+  %0 = load i32, ptr %n.addr, align 4
+  %cmp = icmp sle i32 %0, 1
+  br i1 %cmp, label %cond.true, label %cond.false
+
+cond.true:                                        ; preds = %entry
+  br label %cond.end
+
+cond.false:                                       ; preds = %entry
+  %1 = load i32, ptr %n.addr, align 4
+  %sub = sub nsw i32 %1, 1
+  %call = call signext i32 @fib(i32 noundef signext %sub)
+  %2 = load i32, ptr %n.addr, align 4
+  %sub1 = sub nsw i32 %2, 2
+  %call2 = call signext i32 @fib(i32 noundef signext %sub1)
+  %add = add nsw i32 %call, %call2
+  br label %cond.end
+
+cond.end:                                         ; preds = %cond.false, %cond.true
+  %cond = phi i32 [ 1, %cond.true ], [ %add, %cond.false ]
+  ret i32 %cond
+}
+
+attributes #0 = { "instrument-function-entry-inlined"="mcount" }

llvmbot · 2025-04-15T09:43:07Z

@llvm/pr-subscribers-llvm-transforms

Author: Dominik Steenken (dominik-steenken)

Changes

When compiling with -pg, the EntryExitInstrumenterPass will insert calls to the glibc function mcount at the begining of each MachineFunction.

On SystemZ, these calls require special handling:

The call to mcount needs to happen at the beginning of the prologue.
Prior to the call to mcount, register %r14, the return address of the callee function, must be stored 8 bytes above the stack pointer %r15. After the call to mcount returns, that register needs to be restored.

This commit adds some special handling to the EntryExitInstrumenterPass that keeps the insertion of the mcount function into the module, but skips over insertion of the actual call in order to perform this insertion in the emitPrologue function. There, a simple sequence of store/call/load is inserted, which implements the above.

Noe that the desired change in the EntryExitInstrumenterPass necessitated a change to the signature of the insertCall function, since there now is a possibility that no call is inserted, and thus that needs to be communicated to the caller. We can't ignore that fact either, because we need Changed to be false in the caller for SystemZ, so we can use it in a guard to prevent the deletion of the mcount attribute from the function, which we still need for our postprocessing, aka call insertion, in emitPrologue.

This PR should address issue #121137 . The test inserted here is derived from the example given in that issue.

Full diff: https://github.com/llvm/llvm-project/pull/135767.diff

3 Files Affected:

(modified) llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp (+28)
(modified) llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp (+10-6)
(added) llvm/test/CodeGen/SystemZ/mcount.ll (+35)

diff --git a/llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp b/llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp
index 9561ea544b270..0936690dafdcd 100644
--- a/llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp
+++ b/llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp
@@ -7,6 +7,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "SystemZFrameLowering.h"
+#include "MCTargetDesc/SystemZMCTargetDesc.h"
 #include "SystemZCallingConv.h"
 #include "SystemZInstrInfo.h"
 #include "SystemZMachineFunctionInfo.h"
@@ -18,6 +19,8 @@
 #include "llvm/CodeGen/RegisterScavenging.h"
 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
 #include "llvm/IR/Function.h"
+#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Module.h"
 #include "llvm/Target/TargetMachine.h"
 
 using namespace llvm;
@@ -558,6 +561,31 @@ void SystemZELFFrameLowering::emitPrologue(MachineFunction &MF,
   // to determine the end of the prologue.
   DebugLoc DL;
 
+  // Add mcount instrumentation if necessary.
+  if (MF.getFunction().getFnAttribute("instrument-function-entry-inlined").getValueAsString() == "mcount") {
+
+    // Store return address 8 bytes above stack pointer.
+    BuildMI(MBB, MBBI, DL, ZII->get(SystemZ::STG))
+        .addReg(SystemZ::R14D)
+        .addReg(SystemZ::R15D)
+        .addImm(8)
+        .addReg(0);
+
+    // Call mcount (Regmask 0 to ensure this will not be moved by the
+    // scheduler.).
+    const uint32_t Mask = 0;
+    BuildMI(MBB, MBBI, DL, ZII->get(SystemZ::CallBRASL))
+        .addGlobalAddress(MF.getFunction().getParent()->getFunction("mcount"))
+        .addRegMask(&Mask);
+
+    // Reload return address drom 8 bytes above stack pointer.
+    BuildMI(MBB, MBBI, DL, ZII->get(SystemZ::LG))
+      .addReg(SystemZ::R14D)
+      .addReg(SystemZ::R15D)
+      .addImm(8)
+      .addReg(0);
+  }
+
   // The current offset of the stack pointer from the CFA.
   int64_t SPOffsetFromCFA = -SystemZMC::ELFCFAOffsetFromInitialSP;
 
diff --git a/llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp b/llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp
index d47f1b4253b54..72e4f8791a91a 100644
--- a/llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp
+++ b/llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp
@@ -22,7 +22,7 @@
 
 using namespace llvm;
 
-static void insertCall(Function &CurFn, StringRef Func,
+static bool insertCall(Function &CurFn, StringRef Func,
                        BasicBlock::iterator InsertionPt, DebugLoc DL) {
   Module &M = *InsertionPt->getParent()->getParent()->getParent();
   LLVMContext &C = InsertionPt->getParent()->getContext();
@@ -63,12 +63,16 @@ static void insertCall(Function &CurFn, StringRef Func,
                                   false));
       CallInst *Call = CallInst::Create(Fn, RetAddr, "", InsertionPt);
       Call->setDebugLoc(DL);
+    } else if (TargetTriple.isSystemZ()) {
+      M.getOrInsertFunction(Func, Type::getVoidTy(C));
+      // skip insertion for `mcount` on SystemZ. This will be handled later in `emitPrologue`.
+      return false;
     } else {
       FunctionCallee Fn = M.getOrInsertFunction(Func, Type::getVoidTy(C));
       CallInst *Call = CallInst::Create(Fn, "", InsertionPt);
       Call->setDebugLoc(DL);
     }
-    return;
+    return true;
   }
 
   if (Func == "__cyg_profile_func_enter" || Func == "__cyg_profile_func_exit") {
@@ -87,7 +91,7 @@ static void insertCall(Function &CurFn, StringRef Func,
     CallInst *Call =
         CallInst::Create(Fn, ArrayRef<Value *>(Args), "", InsertionPt);
     Call->setDebugLoc(DL);
-    return;
+    return true;
   }
 
   // We only know how to call a fixed set of instrumentation functions, because
@@ -129,9 +133,9 @@ static bool runOnFunction(Function &F, bool PostInlining) {
     if (auto SP = F.getSubprogram())
       DL = DILocation::get(SP->getContext(), SP->getScopeLine(), 0, SP);
 
-    insertCall(F, EntryFunc, F.begin()->getFirstInsertionPt(), DL);
-    Changed = true;
-    F.removeFnAttr(EntryAttr);
+    Changed = insertCall(F, EntryFunc, F.begin()->getFirstInsertionPt(), DL);
+    if (Changed)
+      F.removeFnAttr(EntryAttr);
   }
 
   if (!ExitFunc.empty()) {
diff --git a/llvm/test/CodeGen/SystemZ/mcount.ll b/llvm/test/CodeGen/SystemZ/mcount.ll
new file mode 100644
index 0000000000000..01bd34548f125
--- /dev/null
+++ b/llvm/test/CodeGen/SystemZ/mcount.ll
@@ -0,0 +1,35 @@
+; Test proper insertion of mcount instrumentation
+;
+; RUN: llc < %s -mtriple=s390x-linux-gnu -o - | FileCheck %s
+;
+; CHECK: # %bb.0:
+; CHECK-NEXT: stg %r14, 8(%r15)
+; CHECK-NEXT: brasl %r14, mcount@PLT
+; CHECK-NEXT: lg %r14, 8(%r15)
+define dso_local signext i32 @fib(i32 noundef signext %n) #0 {
+entry:
+  %n.addr = alloca i32, align 4
+  store i32 %n, ptr %n.addr, align 4
+  %0 = load i32, ptr %n.addr, align 4
+  %cmp = icmp sle i32 %0, 1
+  br i1 %cmp, label %cond.true, label %cond.false
+
+cond.true:                                        ; preds = %entry
+  br label %cond.end
+
+cond.false:                                       ; preds = %entry
+  %1 = load i32, ptr %n.addr, align 4
+  %sub = sub nsw i32 %1, 1
+  %call = call signext i32 @fib(i32 noundef signext %sub)
+  %2 = load i32, ptr %n.addr, align 4
+  %sub1 = sub nsw i32 %2, 2
+  %call2 = call signext i32 @fib(i32 noundef signext %sub1)
+  %add = add nsw i32 %call, %call2
+  br label %cond.end
+
+cond.end:                                         ; preds = %cond.false, %cond.true
+  %cond = phi i32 [ 1, %cond.true ], [ %add, %cond.false ]
+  ret i32 %cond
+}
+
+attributes #0 = { "instrument-function-entry-inlined"="mcount" }

github-actions · 2025-04-15T09:45:01Z

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

dominik-steenken · 2025-04-16T08:27:54Z

The "Linux Premerge Checks" fail seems procedural rather than related to the PR:

Error: Failed to CreateArtifact: Received non-retryable error: Failed request: (409) Conflict: an artifact with this name already exists on the workflow run

How can i request a rebuild?

dominik-steenken · 2025-04-25T09:23:51Z

@uweigand This is ready for review now.

llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp

llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp

dominik-steenken · 2025-05-02T09:06:45Z

I incorporated the review comments. Will rebase and push in a few moments.

When compiling with `-pg`, the `EntryExitInstrumenterPass` will insert calls to the glibc function `mcount` at the begining of each `MachineFunction`. On SystemZ, these calls require special handling: - The call to `mcount` needs to happen at the beginning of the prologue. - Prior to the call to `mcount`, register `%r14`, the return address of the callee function, must be stored 8 bytes above the stack pointer `%r15`. After the call to `mcount` returns, that register needs to be restored. This commit adds some special handling to the EntryExitInstrumenterPass that keeps the insertion of the mcount function into the module, but skips over insertion of the actual call in order to perform this insertion in the `emitPrologue` function. There, a simple sequence of store/call/load is inserted, which implements the above.

uweigand

LGTM now, thanks!

llvm-ci · 2025-05-02T12:38:58Z

LLVM Buildbot has detected a new failure on builder llvm-clang-x86_64-expensive-checks-debian running on gribozavr4 while building llvm at step 6 "test-build-unified-tree-check-all".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/16/builds/18287

Here is the relevant piece of the build log for the reference

Step 6 (test-build-unified-tree-check-all) failure: test (failure)
******************** TEST 'LLVM :: CodeGen/SystemZ/mcount.ll' FAILED ********************
Exit Code: 2

Command Output (stderr):
--
/b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/llc < /b/1/llvm-clang-x86_64-expensive-checks-debian/llvm-project/llvm/test/CodeGen/SystemZ/mcount.ll -mtriple=s390x-linux-gnu -o - | /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/FileCheck /b/1/llvm-clang-x86_64-expensive-checks-debian/llvm-project/llvm/test/CodeGen/SystemZ/mcount.ll # RUN: at line 3
+ /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/llc -mtriple=s390x-linux-gnu -o -
+ /b/1/llvm-clang-x86_64-expensive-checks-debian/build/bin/FileCheck /b/1/llvm-clang-x86_64-expensive-checks-debian/llvm-project/llvm/test/CodeGen/SystemZ/mcount.ll

# After Prologue/Epilogue Insertion & Frame Finalization
# Machine code for function fib: NoPHIs, TracksLiveness, NoVRegs, TiedOpsRewritten, TracksDebugUserValues
Frame Objects:
  fi#-4: size=8, align=8, fixed, at location [SP-160]
  fi#-3: size=8, align=8, fixed, at location [SP-40]
  fi#-2: size=8, align=8, fixed, at location [SP-48]
  fi#-1: size=8, align=8, fixed, at location [SP-56]
  fi#0: size=4, align=4, at location [SP-164]
Function Live Ins: $r2d

bb.0.entry:
  successors: %bb.2(0x40000000), %bb.1(0x40000000); %bb.2(50.00%), %bb.1(50.00%)
  liveins: $r2d, $r13d, $r15d, $r14d
  STG $r14d, $r15d, 8, $noreg
  CallBRASL @mcount, <regmask $f0d $f1d $f2d $f3d $f4d $f5d $f6d $f7d $f8d $f9d $f10d $f11d $f12d $f13d $f14d $f15d $f0h $f1h $f2h $f3h $f4h $f5h $f6h $f7h $f8h $f9h $f10h $f11h $f12h $f13h $f14h $f15h $f0q and 72 more...>, implicit-def $r14d, implicit-def $cc, implicit $fpc
  LG $r14d, $r15d, 8, $noreg
  STMG killed $r13d, killed $r15d, $r15d, 104, implicit killed $r14d
  CFI_INSTRUCTION offset $r13d, -56
  CFI_INSTRUCTION offset $r14d, -48
  CFI_INSTRUCTION offset $r15d, -40
  $r15d = AGHI $r15d(tied-def 0), -168, implicit-def dead $cc
  CFI_INSTRUCTION def_cfa_offset 328
  ST renamable $r2l, $r15d, 164, $noreg :: (store (s32) into %ir.n.addr)
  CHI renamable $r2l, 2, implicit-def $cc, implicit killed $r2d
  renamable $r2l = LHI 1, implicit-def $r2d
  BRC 14, 4, %bb.2, implicit killed $cc
  J %bb.1

bb.1.cond.false:
; predecessors: %bb.0
  successors: %bb.2(0x80000000); %bb.2(100.00%)

  renamable $r0l = LHI -1
  renamable $r0l = A killed renamable $r0l(tied-def 0), $r15d, 164, $noreg, implicit-def dead $cc :: (dereferenceable load (s32) from %ir.n.addr)
  renamable $r2d = LGFR killed renamable $r0l
  CallBRASL @fib, $r2d, <regmask $f8d $f9d $f10d $f11d $f12d $f13d $f14d $f15d $f8h $f9h $f10h $f11h $f12h $f13h $f14h $f15h $f8q $f9q $f12q $f13q $f8s $f9s $f10s $f11s $f12s $f13s $f14s $f15s $r6d $r7d $r8d $r9d $r10d and 30 more...>, implicit-def dead $r14d, implicit-def dead $cc, implicit $fpc, implicit-def $r2d
  renamable $r0l = LHI -2
  renamable $r0l = A killed renamable $r0l(tied-def 0), $r15d, 164, $noreg, implicit-def dead $cc :: (dereferenceable load (s32) from %ir.n.addr)
  renamable $r0d = LGFR killed renamable $r0l
  renamable $r13d = COPY $r2d
  $r2d = COPY killed renamable $r0d
...

dominik-steenken · 2025-05-02T12:43:57Z

checking...

Commit `083b4a3d66` introduced a store-and-load pair around the `BRASL` call to mcount. That load instruction did not properly declare it's target register as defined, leading to a bad machine instruction. This commit fixes this by explicitly labeling `%r14` on the load as `def`.

Commit `083b4a3d66` introduced a store-and-load pair around the `BRASL` call to mcount. That load instruction did not properly declare its target register as defined, leading to a bad machine instruction. This commit fixes this by explicitly labeling `%r14` on the load as `def`.

RKSimon · 2025-05-02T17:57:04Z

@dominik-steenken I'm still seeing an EXPENSIVE_CHECKS build failure - any luck tracking it down please?

dominik-steenken · 2025-05-02T18:30:26Z

Will look into it shortly

dominik-steenken · 2025-05-02T19:41:55Z

@RKSimon I'm not exactly sure what i'm looking at with this sanitizer test fail. I'm currently trying to reproduce it locally. However, looking at that buildbot, it seems the next build is fine, so i'm cautiously optimistic that this was some sort of transient failure.

RKSimon · 2025-05-02T19:54:20Z

I've only seen this on an EXPENSIVE_CHECKS build so far. You need to enable the LLVM_ENABLE_EXPENSIVE_CHECKS cmake def

When compiling with `-pg`, the `EntryExitInstrumenterPass` will insert calls to the glibc function `mcount` at the begining of each `MachineFunction`. On SystemZ, these calls require special handling: - The call to `mcount` needs to happen at the beginning of the prologue. - Prior to the call to `mcount`, register `%r14`, the return address of the callee function, must be stored 8 bytes above the stack pointer `%r15`. After the call to `mcount` returns, that register needs to be restored. This commit adds some special handling to the EntryExitInstrumenterPass that keeps the insertion of the mcount function into the module, but skips over insertion of the actual call in order to perform this insertion in the `emitPrologue` function. There, a simple sequence of store/call/load is inserted, which implements the above. The desired change in the `EntryExitInstrumenterPass` necessitated the addition of a new attribute and attribute kind to each function, which is used to trigger the postprocessing, aka call insertion, in `emitPrologue`. Note that the new attribute must be of a different kind than the `mcount` atribute, since otherwise it would replace that attribute and later be deleted by the code that intended to delete `mcount`. The new attribnute is called `insert-mcount`, while the attribute kind is `systemz-backend`, to clearly mark it as a SystemZ-specific backend concern. This PR should address issue llvm#121137 . The test inserted here is derived from the example given in that issue.

Commit `083b4a3d66` introduced a store-and-load pair around the `BRASL` call to mcount. That load instruction did not properly declare its target register as defined, leading to a bad machine instruction. This commit fixes this by explicitly labeling `%r14` on the load as `def`.

When compiling with `-pg`, the `EntryExitInstrumenterPass` will insert calls to the glibc function `mcount` at the begining of each `MachineFunction`. On SystemZ, these calls require special handling: - The call to `mcount` needs to happen at the beginning of the prologue. - Prior to the call to `mcount`, register `%r14`, the return address of the callee function, must be stored 8 bytes above the stack pointer `%r15`. After the call to `mcount` returns, that register needs to be restored. This commit adds some special handling to the EntryExitInstrumenterPass that keeps the insertion of the mcount function into the module, but skips over insertion of the actual call in order to perform this insertion in the `emitPrologue` function. There, a simple sequence of store/call/load is inserted, which implements the above. The desired change in the `EntryExitInstrumenterPass` necessitated the addition of a new attribute and attribute kind to each function, which is used to trigger the postprocessing, aka call insertion, in `emitPrologue`. Note that the new attribute must be of a different kind than the `mcount` atribute, since otherwise it would replace that attribute and later be deleted by the code that intended to delete `mcount`. The new attribnute is called `insert-mcount`, while the attribute kind is `systemz-backend`, to clearly mark it as a SystemZ-specific backend concern. This PR should address issue llvm#121137 . The test inserted here is derived from the example given in that issue.

Commit `083b4a3d66` introduced a store-and-load pair around the `BRASL` call to mcount. That load instruction did not properly declare its target register as defined, leading to a bad machine instruction. This commit fixes this by explicitly labeling `%r14` on the load as `def`.

When compiling with `-pg`, the `EntryExitInstrumenterPass` will insert calls to the glibc function `mcount` at the begining of each `MachineFunction`. On SystemZ, these calls require special handling: - The call to `mcount` needs to happen at the beginning of the prologue. - Prior to the call to `mcount`, register `%r14`, the return address of the callee function, must be stored 8 bytes above the stack pointer `%r15`. After the call to `mcount` returns, that register needs to be restored. This commit adds some special handling to the EntryExitInstrumenterPass that keeps the insertion of the mcount function into the module, but skips over insertion of the actual call in order to perform this insertion in the `emitPrologue` function. There, a simple sequence of store/call/load is inserted, which implements the above. The desired change in the `EntryExitInstrumenterPass` necessitated the addition of a new attribute and attribute kind to each function, which is used to trigger the postprocessing, aka call insertion, in `emitPrologue`. Note that the new attribute must be of a different kind than the `mcount` atribute, since otherwise it would replace that attribute and later be deleted by the code that intended to delete `mcount`. The new attribnute is called `insert-mcount`, while the attribute kind is `systemz-backend`, to clearly mark it as a SystemZ-specific backend concern. This PR should address issue llvm#121137 . The test inserted here is derived from the example given in that issue.

Commit `083b4a3d66` introduced a store-and-load pair around the `BRASL` call to mcount. That load instruction did not properly declare its target register as defined, leading to a bad machine instruction. This commit fixes this by explicitly labeling `%r14` on the load as `def`.

When compiling with `-pg`, the `EntryExitInstrumenterPass` will insert calls to the glibc function `mcount` at the begining of each `MachineFunction`. On SystemZ, these calls require special handling: - The call to `mcount` needs to happen at the beginning of the prologue. - Prior to the call to `mcount`, register `%r14`, the return address of the callee function, must be stored 8 bytes above the stack pointer `%r15`. After the call to `mcount` returns, that register needs to be restored. This commit adds some special handling to the EntryExitInstrumenterPass that keeps the insertion of the mcount function into the module, but skips over insertion of the actual call in order to perform this insertion in the `emitPrologue` function. There, a simple sequence of store/call/load is inserted, which implements the above. The desired change in the `EntryExitInstrumenterPass` necessitated the addition of a new attribute and attribute kind to each function, which is used to trigger the postprocessing, aka call insertion, in `emitPrologue`. Note that the new attribute must be of a different kind than the `mcount` atribute, since otherwise it would replace that attribute and later be deleted by the code that intended to delete `mcount`. The new attribnute is called `insert-mcount`, while the attribute kind is `systemz-backend`, to clearly mark it as a SystemZ-specific backend concern. This PR should address issue llvm#121137 . The test inserted here is derived from the example given in that issue.

Commit `083b4a3d66` introduced a store-and-load pair around the `BRASL` call to mcount. That load instruction did not properly declare its target register as defined, leading to a bad machine instruction. This commit fixes this by explicitly labeling `%r14` on the load as `def`.

llvmbot added backend:SystemZ llvm:transforms labels Apr 15, 2025

dominik-steenken force-pushed the fix-mcount-v2 branch from 296fa31 to dd52808 Compare April 15, 2025 11:00

dominik-steenken force-pushed the fix-mcount-v2 branch 2 times, most recently from c246c4e to d715e01 Compare April 25, 2025 09:20

dominik-steenken force-pushed the fix-mcount-v2 branch from d715e01 to 955a04b Compare April 25, 2025 09:26

uweigand reviewed Apr 25, 2025

View reviewed changes

llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp Outdated Show resolved Hide resolved

llvm/lib/Target/SystemZ/SystemZFrameLowering.cpp Outdated Show resolved Hide resolved

llvm/lib/Transforms/Utils/EntryExitInstrumenter.cpp Outdated Show resolved Hide resolved

dominik-steenken force-pushed the fix-mcount-v2 branch from 955a04b to f7ea728 Compare May 2, 2025 09:13

Implement review suggestions

77f2ba0

dominik-steenken force-pushed the fix-mcount-v2 branch from f7ea728 to 77f2ba0 Compare May 2, 2025 09:18

Comment typos.

89421e5

uweigand approved these changes May 2, 2025

View reviewed changes

uweigand merged commit 083b4a3 into llvm:main May 2, 2025
6 of 10 checks passed

uweigand mentioned this pull request May 2, 2025

SystemZ: Missing gprof support #121137

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[SystemZ] Add proper mcount handling #135767

[SystemZ] Add proper mcount handling #135767

dominik-steenken commented Apr 15, 2025 •

edited

Loading

llvmbot commented Apr 15, 2025

llvmbot commented Apr 15, 2025

github-actions bot commented Apr 15, 2025 •

edited

Loading

dominik-steenken commented Apr 16, 2025

dominik-steenken commented Apr 25, 2025

dominik-steenken commented May 2, 2025 •

edited

Loading

uweigand left a comment

llvm-ci commented May 2, 2025

dominik-steenken commented May 2, 2025

RKSimon commented May 2, 2025

dominik-steenken commented May 2, 2025

dominik-steenken commented May 2, 2025

RKSimon commented May 2, 2025

[SystemZ] Add proper mcount handling #135767

[SystemZ] Add proper mcount handling #135767

Conversation

dominik-steenken commented Apr 15, 2025 • edited Loading

llvmbot commented Apr 15, 2025

llvmbot commented Apr 15, 2025

github-actions bot commented Apr 15, 2025 • edited Loading

dominik-steenken commented Apr 16, 2025

dominik-steenken commented Apr 25, 2025

dominik-steenken commented May 2, 2025 • edited Loading

uweigand left a comment

Choose a reason for hiding this comment

llvm-ci commented May 2, 2025

dominik-steenken commented May 2, 2025

RKSimon commented May 2, 2025

dominik-steenken commented May 2, 2025

dominik-steenken commented May 2, 2025

RKSimon commented May 2, 2025

dominik-steenken commented Apr 15, 2025 •

edited

Loading

github-actions bot commented Apr 15, 2025 •

edited

Loading

dominik-steenken commented May 2, 2025 •

edited

Loading