[BPF] Don't crash on missing line info

[tamird]

Summary

When compiling Rust code we may end up with calls to functions provided
by other code units. Presently this code crashes on a null pointer
dereference - this patch avoids that crash and adds a test.

[tamird]

Avoid the crash

[tamird]

Revert back to better error

[eddyz87]

This error could be detected using the following test case:

$ cat test2.c
; ModuleID = 'test2.c'
source_filename = "test2.c"
target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n32:64-S128"
target triple = "bpf"

; Function Attrs: noinline nounwind optnone
define dso_local i64 @foo() {
entry:
  ret i64 42
}

!llvm.dbg.cu = !{!0}
!llvm.module.flags = !{!2, !3, !4, !5}

!0 = distinct !DICompileUnit(language: DW_LANG_C11, file: !1, producer: "foo", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: false, nameTableKind: None)
!1 = !DIFile(filename: "file.c", directory: "/some/dir", checksumkind: CSK_MD5, checksum: "f58a7abbd9253986370acb1013fc9e55")
!2 = !{i32 7, !"Dwarf Version", i32 5}
!3 = !{i32 2, !"Debug Info Version", i32 3}
!4 = !{i32 1, !"wchar_size", i32 4}
!5 = !{i32 7, !"frame-pointer", i32 2}
!6 = !{!"clang version 18.0.0 (/home/eddy/work/llvm-project/clang 61cd4a1d35aa7c0f7b312b8f43ed3bff33f20d55)"}

$ llc test2.ll -o /dev/null
/home/eddy/work/llvm-project/llvm/lib/Target/BPF/BTFDebug.cpp:1640:42: runtime error: member call on null pointer of type 'llvm::DISubprogram'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior /home/eddy/work/llvm-project/llvm/lib/Target/BPF/BTFDebug.cpp:1640:42 in 

Necessary conditions are: module with debug attributes (llvm.dbg.cu -> DICompileUnit) and a function and instruction w/o debug metadata. I think such test case should be added to the suite.
Also, I don't see a point in reporting this (either as warning or as error). Absence of DISubprogram is ignored in BTFDebug::processFuncPrototypes and it two other places getSubprogram() can't return null. I think the following patch should suffice:

diff --git a/llvm/lib/Target/BPF/BTFDebug.cpp b/llvm/lib/Target/BPF/BTFDebug.cpp
index d886edaff555..6df421934b70 100644
--- a/llvm/lib/Target/BPF/BTFDebug.cpp
+++ b/llvm/lib/Target/BPF/BTFDebug.cpp
@@ -1634,8 +1635,8 @@ void BTFDebug::beginInstruction(const MachineInstr *MI) {
   if (!DL || PrevInstLoc == DL) {
     // This instruction will be skipped, no LineInfo has
     // been generated, construct one based on function signature.
-    if (LineInfoGenerated == false) {
-      auto *S = MI->getMF()->getFunction().getSubprogram();
+    DISubprogram *S = MI->getMF()->getFunction().getSubprogram();
+    if (!LineInfoGenerated && S) {
       MCSymbol *FuncLabel = Asm->getFunctionBegin();
       constructLineInfo(S, FuncLabel, S->getLine(), 0);
       LineInfoGenerated = true;

[tamird]

In D156446#4554118, @eddyz87 wrote:

This error could be detected using the following test case:

$ cat test2.c
; ModuleID = 'test2.c'
source_filename = "test2.c"
target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n32:64-S128"
target triple = "bpf"

; Function Attrs: noinline nounwind optnone
define dso_local i64 @foo() {
entry:
  ret i64 42
}

!llvm.dbg.cu = !{!0}
!llvm.module.flags = !{!2, !3, !4, !5}

!0 = distinct !DICompileUnit(language: DW_LANG_C11, file: !1, producer: "foo", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, splitDebugInlining: false, nameTableKind: None)
!1 = !DIFile(filename: "file.c", directory: "/some/dir", checksumkind: CSK_MD5, checksum: "f58a7abbd9253986370acb1013fc9e55")
!2 = !{i32 7, !"Dwarf Version", i32 5}
!3 = !{i32 2, !"Debug Info Version", i32 3}
!4 = !{i32 1, !"wchar_size", i32 4}
!5 = !{i32 7, !"frame-pointer", i32 2}
!6 = !{!"clang version 18.0.0 (/home/eddy/work/llvm-project/clang 61cd4a1d35aa7c0f7b312b8f43ed3bff33f20d55)"}

$ llc test2.ll -o /dev/null
/home/eddy/work/llvm-project/llvm/lib/Target/BPF/BTFDebug.cpp:1640:42: runtime error: member call on null pointer of type 'llvm::DISubprogram'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior /home/eddy/work/llvm-project/llvm/lib/Target/BPF/BTFDebug.cpp:1640:42 in 

Necessary conditions are: module with debug attributes (llvm.dbg.cu -> DICompileUnit) and a function and instruction w/o debug metadata. I think such test case should be added to the suite.
Also, I don't see a point in reporting this (either as warning or as error). Absence of DISubprogram is ignored in BTFDebug::processFuncPrototypes and it two other places getSubprogram() can't return null. I think the following patch should suffice:

diff --git a/llvm/lib/Target/BPF/BTFDebug.cpp b/llvm/lib/Target/BPF/BTFDebug.cpp
index d886edaff555..6df421934b70 100644
--- a/llvm/lib/Target/BPF/BTFDebug.cpp
+++ b/llvm/lib/Target/BPF/BTFDebug.cpp
@@ -1634,8 +1635,8 @@ void BTFDebug::beginInstruction(const MachineInstr *MI) {
   if (!DL || PrevInstLoc == DL) {
     // This instruction will be skipped, no LineInfo has
     // been generated, construct one based on function signature.
-    if (LineInfoGenerated == false) {
-      auto *S = MI->getMF()->getFunction().getSubprogram();
+    DISubprogram *S = MI->getMF()->getFunction().getSubprogram();
+    if (!LineInfoGenerated && S) {
       MCSymbol *FuncLabel = Asm->getFunctionBegin();
       constructLineInfo(S, FuncLabel, S->getLine(), 0);
       LineInfoGenerated = true;

Sounds good! I'll do that. Can you share the C source code of the program as well please?

[eddyz87]

In D156446#4554385, @tamird wrote:

Sounds good! I'll do that. Can you share the C source code of the program as well please?

// clang -g --target=bpf -S -emit-llvm test2.c -o -
unsigned long foo(void) {
  return 42;
}

Modified to remove !dbg from function and ret instructions.

[tamird]

Avoid crash, add test.

[tamird]

@ast @yonghong-song please review. This replaces https://reviews.llvm.org/D134533.

[ast]

llvm/lib/Target/BPF/BTFDebug.cpp

1379

parens

[tamird]

Remove braces

[tamird]

llvm/lib/Target/BPF/BTFDebug.cpp

1379

Done.

[tamird]

Less const

[ast]

lgtm assuming all tests are green.

[tamird]

Remove one CHECK line.

llvm/lib/Target/BPF/BTFDebug.cpp

for (line_iterator I(*Buf, false), E; I != E; ++I)
       Content.push_back(std::string(*I));
 
   FileContent[FileName] = Content;
   return FileName;
 }
 
 void BTFDebug::constructLineInfo(const DIFile *File, MCSymbol *Label,
                                  uint32_t Line, uint32_t Column) {
-  std::string FileName = populateFileContent(File);
   BTFLineInfo LineInfo;
 
   LineInfo.Label = Label;
+  LineInfo.FileNameOff = 0;
+  LineInfo.LineOff = 0;
+  if (File) {
+    std::string FileName = populateFileContent(File);
     LineInfo.FileNameOff = addString(FileName);
     // If file content is not available, let LineOff = 0.
     if (Line < FileContent[FileName].size())
       LineInfo.LineOff = addString(FileContent[FileName][Line]);
-  else
-    LineInfo.LineOff = 0;
+  }
   LineInfo.LineNum = Line;
   LineInfo.ColumnNum = Column;
   LineInfoTable[SecNameOff].push_back(LineInfo);
 }
 
 void BTFDebug::emitCommonHeader() {
   OS.AddComment("0x" + Twine::utohexstr(BTF::MAGIC));
   OS.emitIntValue(BTF::MAGIC, 2);
void BTFDebug::beginInstruction(const MachineInstr *MI) {
 
   // Skip this instruction if no DebugLoc or the DebugLoc
   // is the same as the previous instruction.
   const DebugLoc &DL = MI->getDebugLoc();
   if (!DL || PrevInstLoc == DL) {
     // This instruction will be skipped, no LineInfo has
     // been generated, construct one based on function signature.
     if (LineInfoGenerated == false) {
-      auto *S = MI->getMF()->getFunction().getSubprogram();
+      const Function &F = MI->getMF()->getFunction();
+      auto *S = F.getSubprogram();
       MCSymbol *FuncLabel = Asm->getFunctionBegin();
+      // The callee may not have been compiled with debug info.
+      if (S) {
         constructLineInfo(S->getFile(), FuncLabel, S->getLine(), 0);
+      } else {
+        constructLineInfo(nullptr, FuncLabel, 0, 0);
+      }
       LineInfoGenerated = true;
     }
 
     return;
   }
 
   // Create a temporary label to remember the insn for lineinfo.
   MCSymbol *LineSym = OS.getContext().createTempSymbol();