Merge branch 'master' into request-struct

Add support for userspace+kernelspace stack trace example using blazesym

pythonbpf/allocation_pass.py

@@ -7,6 +7,7 @@ from pythonbpf.helper import HelperHandlerRegistry
 from pythonbpf.vmlinux_parser.dependency_node import Field
 from .expr import VmlinuxHandlerRegistry
 from pythonbpf.type_deducer import ctypes_to_ir
+from pythonbpf.maps import BPFMapType
 
 logger = logging.getLogger(__name__)
 
@@ -25,7 +26,9 @@ def create_targets_and_rvals(stmt):
     return stmt.targets, [stmt.value]
 
 
-def handle_assign_allocation(builder, stmt, local_sym_tab, structs_sym_tab):
+def handle_assign_allocation(
+    builder, stmt, local_sym_tab, map_sym_tab, structs_sym_tab
+):
     """Handle memory allocation for assignment statements."""
 
     logger.info(f"Handling assignment for allocation: {ast.dump(stmt)}")
@@ -55,7 +58,9 @@ def handle_assign_allocation(builder, stmt, local_sym_tab, structs_sym_tab):
 
         # Determine type and allocate based on rval
         if isinstance(rval, ast.Call):
-            _allocate_for_call(builder, var_name, rval, local_sym_tab, structs_sym_tab)
+            _allocate_for_call(
+                builder, var_name, rval, local_sym_tab, map_sym_tab, structs_sym_tab
+            )
         elif isinstance(rval, ast.Constant):
             _allocate_for_constant(builder, var_name, rval, local_sym_tab)
         elif isinstance(rval, ast.BinOp):
@@ -74,7 +79,9 @@ def handle_assign_allocation(builder, stmt, local_sym_tab, structs_sym_tab):
             )
 
 
-def _allocate_for_call(builder, var_name, rval, local_sym_tab, structs_sym_tab):
+def _allocate_for_call(
+    builder, var_name, rval, local_sym_tab, map_sym_tab, structs_sym_tab
+):
     """Allocate memory for variable assigned from a call."""
 
     if isinstance(rval.func, ast.Name):
@@ -116,15 +123,74 @@ def _allocate_for_call(builder, var_name, rval, local_sym_tab, structs_sym_tab):
 
     elif isinstance(rval.func, ast.Attribute):
         # Map method calls - need double allocation for ptr handling
-        _allocate_for_map_method(builder, var_name, local_sym_tab)
+        _allocate_for_map_method(
+            builder, var_name, rval, local_sym_tab, map_sym_tab, structs_sym_tab
+        )
 
     else:
         logger.warning(f"Unsupported call function type for {var_name}")
 
 
-def _allocate_for_map_method(builder, var_name, local_sym_tab):
+def _allocate_for_map_method(
+    builder, var_name, rval, local_sym_tab, map_sym_tab, structs_sym_tab
+):
     """Allocate memory for variable assigned from map method (double alloc)."""
 
+    map_name = rval.func.value.id
+    method_name = rval.func.attr
+
+    # NOTE: We will have to special case HashMap.lookup which returns a pointer to value type
+    # The value type can be a struct as well, so we need to handle that properly
+    # This special casing is not ideal, as over time other map methods may need similar handling
+    # But for now, we will just handle lookup specifically
+    if map_name not in map_sym_tab:
+        logger.error(f"Map '{map_name}' not found for allocation")
+        return
+
+    if method_name != "lookup":
+        # Fallback allocation for other map methods
+        _allocate_for_map_method_fallback(builder, var_name, local_sym_tab)
+        return
+
+    map_params = map_sym_tab[map_name].params
+    if map_params["type"] != BPFMapType.HASH:
+        logger.warning(
+            "Map method lookup used on non-hash map, using fallback allocation"
+        )
+        _allocate_for_map_method_fallback(builder, var_name, local_sym_tab)
+        return
+
+    value_type = map_params["value"]
+    # Determine IR type for value
+    if isinstance(value_type, str) and value_type in structs_sym_tab:
+        struct_info = structs_sym_tab[value_type]
+        value_ir_type = struct_info.ir_type
+    else:
+        value_ir_type = ctypes_to_ir(value_type)
+
+    if value_ir_type is None:
+        logger.warning(
+            f"Could not determine IR type for map value '{value_type}', using fallback allocation"
+        )
+        _allocate_for_map_method_fallback(builder, var_name, local_sym_tab)
+        return
+
+    # Main variable (pointer to pointer)
+    ir_type = ir.PointerType(ir.IntType(64))
+    var = builder.alloca(ir_type, name=var_name)
+    local_sym_tab[var_name] = LocalSymbol(var, ir_type)
+    # Temporary variable for computed values
+    tmp_ir_type = value_ir_type
+    var_tmp = builder.alloca(tmp_ir_type, name=f"{var_name}_tmp")
+    local_sym_tab[f"{var_name}_tmp"] = LocalSymbol(var_tmp, tmp_ir_type)
+    logger.info(
+        f"Pre-allocated {var_name} and {var_name}_tmp for map method lookup of type {value_ir_type}"
+    )
+
+
+def _allocate_for_map_method_fallback(builder, var_name, local_sym_tab):
+    """Fallback allocation for map method variable (i64* and i64**)."""
+
     # Main variable (pointer to pointer)
     ir_type = ir.PointerType(ir.IntType(64))
     var = builder.alloca(ir_type, name=var_name)
@@ -135,7 +201,9 @@ def _allocate_for_map_method(builder, var_name, local_sym_tab):
     var_tmp = builder.alloca(tmp_ir_type, name=f"{var_name}_tmp")
     local_sym_tab[f"{var_name}_tmp"] = LocalSymbol(var_tmp, tmp_ir_type)
 
-    logger.info(f"Pre-allocated {var_name} and {var_name}_tmp for map method")
+    logger.info(
+        f"Pre-allocated {var_name} and {var_name}_tmp for map method (fallback)"
+    )
 
 
 def _allocate_for_constant(builder, var_name, rval, local_sym_tab):

pythonbpf/debuginfo/debug_info_generator.py

@@ -49,6 +49,10 @@ class DebugInfoGenerator:
             )
         return self._type_cache[key]
 
+    def get_uint8_type(self) -> Any:
+        """Get debug info for signed 8-bit integer"""
+        return self.get_basic_type("char", 8, dc.DW_ATE_unsigned)
+
     def get_int32_type(self) -> Any:
         """Get debug info for signed 32-bit integer"""
         return self.get_basic_type("int", 32, dc.DW_ATE_signed)

pythonbpf/functions/functions_pass.py

@@ -147,7 +147,9 @@ def allocate_mem(
                 structs_sym_tab,
             )
         elif isinstance(stmt, ast.Assign):
-            handle_assign_allocation(builder, stmt, local_sym_tab, structs_sym_tab)
+            handle_assign_allocation(
+                builder, stmt, local_sym_tab, map_sym_tab, structs_sym_tab
+            )
 
     allocate_temp_pool(builder, max_temps_needed, local_sym_tab)
 

pythonbpf/maps/map_debug_info.py

@@ -1,22 +1,31 @@
+import logging
+from llvmlite import ir
 from pythonbpf.debuginfo import DebugInfoGenerator
 from .map_types import BPFMapType
 
+logger: logging.Logger = logging.getLogger(__name__)
+
 
 def create_map_debug_info(module, map_global, map_name, map_params, structs_sym_tab):
     """Generate debug info metadata for BPF maps HASH and PERF_EVENT_ARRAY"""
     generator = DebugInfoGenerator(module)
-
+    logger.info(f"Creating debug info for map {map_name} with params {map_params}")
     uint_type = generator.get_uint32_type()
-    ulong_type = generator.get_uint64_type()
     array_type = generator.create_array_type(
         uint_type, map_params.get("type", BPFMapType.UNSPEC).value
     )
     type_ptr = generator.create_pointer_type(array_type, 64)
     key_ptr = generator.create_pointer_type(
-        array_type if "key_size" in map_params else ulong_type, 64
+        array_type
+        if "key_size" in map_params
+        else _get_key_val_dbg_type(map_params.get("key"), generator, structs_sym_tab),
+        64,
     )
     value_ptr = generator.create_pointer_type(
-        array_type if "value_size" in map_params else ulong_type, 64
+        array_type
+        if "value_size" in map_params
+        else _get_key_val_dbg_type(map_params.get("value"), generator, structs_sym_tab),
+        64,
     )
 
     elements_arr = []
@@ -97,3 +106,65 @@ def create_ringbuf_debug_info(
     )
     map_global.set_metadata("dbg", global_var)
     return global_var
+
+
+def _get_key_val_dbg_type(name, generator, structs_sym_tab):
+    """Get the debug type for key/value based on type object"""
+
+    if not name:
+        logger.warn("No name provided for key/value type, defaulting to uint64")
+        return generator.get_uint64_type()
+
+    type_obj = structs_sym_tab.get(name)
+    if type_obj:
+        return _get_struct_debug_type(type_obj, generator, structs_sym_tab)
+
+    # Fallback to basic types
+    logger.info(f"No struct named {name}, falling back to basic type")
+
+    # NOTE: Only handling int and long for now
+    if name in ["c_int32", "c_uint32"]:
+        return generator.get_uint32_type()
+
+    # Default fallback for now
+    return generator.get_uint64_type()
+
+
+def _get_struct_debug_type(struct_obj, generator, structs_sym_tab):
+    """Recursively create debug type for struct"""
+    elements_arr = []
+    for fld in struct_obj.fields.keys():
+        fld_type = struct_obj.field_type(fld)
+        if isinstance(fld_type, ir.IntType):
+            if fld_type.width == 32:
+                fld_dbg_type = generator.get_uint32_type()
+            else:
+                # NOTE: Assuming 64-bit for all other int types
+                fld_dbg_type = generator.get_uint64_type()
+        elif isinstance(fld_type, ir.ArrayType):
+            # NOTE: Array types have u8 elements only for now
+            # Debug info generation should fail for other types
+            elem_type = fld_type.element
+            if isinstance(elem_type, ir.IntType) and elem_type.width == 8:
+                char_type = generator.get_uint8_type()
+                fld_dbg_type = generator.create_array_type(char_type, fld_type.count)
+            else:
+                logger.warning(
+                    f"Array element type {str(elem_type)} not supported for debug info, skipping"
+                )
+                continue
+        else:
+            # NOTE: Only handling int and char arrays for now
+            logger.warning(
+                f"Field type {str(fld_type)} not supported for debug info, skipping"
+            )
+            continue
+
+        member = generator.create_struct_member(
+            fld, fld_dbg_type, struct_obj.field_size(fld)
+        )
+        elements_arr.append(member)
+    struct_type = generator.create_struct_type(
+        elements_arr, struct_obj.size, is_distinct=True
+    )
+    return struct_type

pythonbpf/maps/maps_pass.py

@@ -48,7 +48,7 @@ def create_bpf_map(module, map_name, map_params):
     map_global.align = 8
 
     logger.info(f"Created BPF map: {map_name} with params {map_params}")
-    return MapSymbol(type=map_params["type"], sym=map_global)
+    return MapSymbol(type=map_params["type"], sym=map_global, params=map_params)
 
 
 def _parse_map_params(rval, expected_args=None):
@@ -105,7 +105,9 @@ def process_ringbuf_map(map_name, rval, module, structs_sym_tab):
     logger.info(f"Ringbuf map parameters: {map_params}")
 
     map_global = create_bpf_map(module, map_name, map_params)
-    create_ringbuf_debug_info(module, map_global.sym, map_name, map_params)
+    create_ringbuf_debug_info(
+        module, map_global.sym, map_name, map_params, structs_sym_tab
+    )
     return map_global
 
 
@@ -119,7 +121,7 @@ def process_hash_map(map_name, rval, module, structs_sym_tab):
     logger.info(f"Map parameters: {map_params}")
     map_global = create_bpf_map(module, map_name, map_params)
     # Generate debug info for BTF
-    create_map_debug_info(module, map_global.sym, map_name, map_params)
+    create_map_debug_info(module, map_global.sym, map_name, map_params, structs_sym_tab)
     return map_global
 
 

pythonbpf/maps/maps_utils.py

@@ -11,6 +11,7 @@ class MapSymbol:
 
     type: BPFMapType
     sym: ir.GlobalVariable
+    params: dict[str, Any] | None = None
 
 
 class MapProcessorRegistry:

pythonbpf/vmlinux_parser/class_handler.py

@@ -16,6 +16,33 @@ def get_module_symbols(module_name: str):
     return [name for name in dir(imported_module)], imported_module
 
 
+def unwrap_pointer_type(type_obj: Any) -> Any:
+    """
+    Recursively unwrap all pointer layers to get the base type.
+
+    This handles multiply nested pointers like LP_LP_struct_attribute_group
+    and returns the base type (struct_attribute_group).
+
+    Stops unwrapping when reaching a non-pointer type (one without _type_ attribute).
+
+    Args:
+        type_obj: The type object to unwrap
+
+    Returns:
+        The base type after unwrapping all pointer layers
+    """
+    current_type = type_obj
+    # Keep unwrapping while it's a pointer/array type (has _type_)
+    # But stop if _type_ is just a string or basic type marker
+    while hasattr(current_type, "_type_"):
+        next_type = current_type._type_
+        # Stop if _type_ is a string (like 'c' for c_char)
+        if isinstance(next_type, str):
+            break
+        current_type = next_type
+    return current_type
+
+
 def process_vmlinux_class(
     node,
     llvm_module,
@@ -158,13 +185,90 @@ def process_vmlinux_post_ast(
                         if hasattr(elem_type, "_length_") and is_complex_type:
                             type_length = elem_type._length_
 
-                        if containing_type.__module__ == "vmlinux":
-                            new_dep_node.add_dependent(
-                                elem_type._type_.__name__
-                                if hasattr(elem_type._type_, "__name__")
-                                else str(elem_type._type_)
+                        # Unwrap all pointer layers to get the base type for dependency tracking
+                        base_type = unwrap_pointer_type(elem_type)
+                        base_type_module = getattr(base_type, "__module__", None)
+
+                        if base_type_module == "vmlinux":
+                            base_type_name = (
+                                base_type.__name__
+                                if hasattr(base_type, "__name__")
+                                else str(base_type)
+                            )
+                            # ONLY add vmlinux types as dependencies
+                            new_dep_node.add_dependent(base_type_name)
+
+                            logger.debug(
+                                f"{containing_type} containing type of parent {elem_name} with {elem_type} and ctype {ctype_complex_type} and length {type_length}"
+                            )
+                            new_dep_node.set_field_containing_type(
+                                elem_name, containing_type
+                            )
+                            new_dep_node.set_field_type_size(elem_name, type_length)
+                            new_dep_node.set_field_ctype_complex_type(
+                                elem_name, ctype_complex_type
+                            )
+                            new_dep_node.set_field_type(elem_name, elem_type)
+
+                            # Check the containing_type module to decide whether to recurse
+                            containing_type_module = getattr(
+                                containing_type, "__module__", None
+                            )
+                            if containing_type_module == "vmlinux":
+                                # Also unwrap containing_type to get base type name
+                                base_containing_type = unwrap_pointer_type(
+                                    containing_type
+                                )
+                                containing_type_name = (
+                                    base_containing_type.__name__
+                                    if hasattr(base_containing_type, "__name__")
+                                    else str(base_containing_type)
+                                )
+
+                                # Check for self-reference or already processed
+                                if containing_type_name == current_symbol_name:
+                                    # Self-referential pointer
+                                    logger.debug(
+                                        f"Self-referential pointer in {current_symbol_name}.{elem_name}"
+                                    )
+                                    new_dep_node.set_field_ready(elem_name, True)
+                                elif handler.has_node(containing_type_name):
+                                    # Already processed
+                                    logger.debug(
+                                        f"Reusing already processed {containing_type_name}"
+                                    )
+                                    new_dep_node.set_field_ready(elem_name, True)
+                                else:
+                                    # Process recursively - use base containing type, not the pointer wrapper
+                                    new_dep_node.add_dependent(containing_type_name)
+                                    process_vmlinux_post_ast(
+                                        base_containing_type,
+                                        llvm_handler,
+                                        handler,
+                                        processing_stack,
+                                    )
+                                    new_dep_node.set_field_ready(elem_name, True)
+                            elif (
+                                containing_type_module == ctypes.__name__
+                                or containing_type_module is None
+                            ):
+                                logger.debug(
+                                    f"Processing ctype internal{containing_type}"
+                                )
+                                new_dep_node.set_field_ready(elem_name, True)
+                            else:
+                                raise TypeError(
+                                    f"Module not supported in recursive resolution: {containing_type_module}"
+                                )
+                        elif (
+                            base_type_module == ctypes.__name__
+                            or base_type_module is None
+                        ):
+                            # Handle ctypes or types with no module (like some internal ctypes types)
+                            # DO NOT add ctypes as dependencies - just set field metadata and mark ready
+                            logger.debug(
+                                f"Base type {base_type} is ctypes - NOT adding as dependency, just processing field"
                             )
-                        elif containing_type.__module__ == ctypes.__name__:
                             if isinstance(elem_type, type):
                                 if issubclass(elem_type, ctypes.Array):
                                     ctype_complex_type = ctypes.Array
@@ -176,57 +280,20 @@ def process_vmlinux_post_ast(
                                     )
                             else:
                                 raise TypeError("Unsupported ctypes subclass")
-                        else:
-                            raise ImportError(
-                                f"Unsupported module of {containing_type}"
-                            )
-                        logger.debug(
-                            f"{containing_type} containing type of parent {elem_name} with {elem_type} and ctype {ctype_complex_type} and length {type_length}"
-                        )
-                        new_dep_node.set_field_containing_type(
-                            elem_name, containing_type
-                        )
-                        new_dep_node.set_field_type_size(elem_name, type_length)
-                        new_dep_node.set_field_ctype_complex_type(
-                            elem_name, ctype_complex_type
-                        )
-                        new_dep_node.set_field_type(elem_name, elem_type)
-                        if containing_type.__module__ == "vmlinux":
-                            containing_type_name = (
-                                containing_type.__name__
-                                if hasattr(containing_type, "__name__")
-                                else str(containing_type)
-                            )
 
-                            # Check for self-reference or already processed
-                            if containing_type_name == current_symbol_name:
-                                # Self-referential pointer
-                                logger.debug(
-                                    f"Self-referential pointer in {current_symbol_name}.{elem_name}"
-                                )
-                                new_dep_node.set_field_ready(elem_name, True)
-                            elif handler.has_node(containing_type_name):
-                                # Already processed
-                                logger.debug(
-                                    f"Reusing already processed {containing_type_name}"
-                                )
-                                new_dep_node.set_field_ready(elem_name, True)
-                            else:
-                                # Process recursively - THIS WAS MISSING
-                                new_dep_node.add_dependent(containing_type_name)
-                                process_vmlinux_post_ast(
-                                    containing_type,
-                                    llvm_handler,
-                                    handler,
-                                    processing_stack,
-                                )
-                                new_dep_node.set_field_ready(elem_name, True)
-                        elif containing_type.__module__ == ctypes.__name__:
-                            logger.debug(f"Processing ctype internal{containing_type}")
+                            # Set field metadata but DO NOT add dependency or recurse
+                            new_dep_node.set_field_containing_type(
+                                elem_name, containing_type
+                            )
+                            new_dep_node.set_field_type_size(elem_name, type_length)
+                            new_dep_node.set_field_ctype_complex_type(
+                                elem_name, ctype_complex_type
+                            )
+                            new_dep_node.set_field_type(elem_name, elem_type)
                             new_dep_node.set_field_ready(elem_name, True)
                         else:
-                            raise TypeError(
-                                "Module not supported in recursive resolution"
+                            raise ImportError(
+                                f"Unsupported module of {base_type}: {base_type_module}"
                             )
                     else:
                         new_dep_node.add_dependent(
@@ -245,9 +312,12 @@ def process_vmlinux_post_ast(
                     raise ValueError(
                         f"{elem_name} with type {elem_type} from module {module_name} not supported in recursive resolver"
                     )
-
+    elif module_name == ctypes.__name__ or module_name is None:
+        # Handle ctypes types - these don't need processing, just return
+        logger.debug(f"Skipping ctypes type {current_symbol_name}")
+        return True
     else:
-        raise ImportError("UNSUPPORTED Module")
+        raise ImportError(f"UNSUPPORTED Module {module_name}")
 
     logger.info(
         f"{current_symbol_name} processed and handler readiness {handler.is_ready}"

pythonbpf/vmlinux_parser/import_detector.py

@@ -11,7 +11,9 @@ from .class_handler import process_vmlinux_class
 logger = logging.getLogger(__name__)
 
 
-def detect_import_statement(tree: ast.AST) -> list[tuple[str, ast.ImportFrom]]:
+def detect_import_statement(
+    tree: ast.AST,
+) -> list[tuple[str, ast.ImportFrom, str, str]]:
     """
     Parse AST and detect import statements from vmlinux.
 
@@ -25,7 +27,7 @@ def detect_import_statement(tree: ast.AST) -> list[tuple[str, ast.ImportFrom]]:
         List of tuples containing (module_name, imported_item) for each vmlinux import
 
     Raises:
-        SyntaxError: If multiple imports from vmlinux are attempted or import * is used
+        SyntaxError: If import * is used
     """
     vmlinux_imports = []
 
@@ -40,28 +42,19 @@ def detect_import_statement(tree: ast.AST) -> list[tuple[str, ast.ImportFrom]]:
                         "Please import specific types explicitly."
                     )
 
-                # Check for multiple imports: from vmlinux import A, B, C
-                if len(node.names) > 1:
-                    imported_names = [alias.name for alias in node.names]
-                    raise SyntaxError(
-                        f"Multiple imports from vmlinux are not supported. "
-                        f"Found: {', '.join(imported_names)}. "
-                        f"Please use separate import statements for each type."
-                    )
-
                 # Check if no specific import is specified (should not happen with valid Python)
                 if len(node.names) == 0:
                     raise SyntaxError(
                         "Import from vmlinux must specify at least one type."
                     )
 
-                # Valid single import
+                # Support multiple imports: from vmlinux import A, B, C
                 for alias in node.names:
                     import_name = alias.name
-                    # Use alias if provided, otherwise use the original name (commented)
-                    # as_name = alias.asname if alias.asname else alias.name
-                    vmlinux_imports.append(("vmlinux", node))
-                    logger.info(f"Found vmlinux import: {import_name}")
+                    # Use alias if provided, otherwise use the original name
+                    as_name = alias.asname if alias.asname else alias.name
+                    vmlinux_imports.append(("vmlinux", node, import_name, as_name))
+                    logger.info(f"Found vmlinux import: {import_name} as {as_name}")
 
         # Handle "import vmlinux" statements (not typical but should be rejected)
         elif isinstance(node, ast.Import):
@@ -103,40 +96,37 @@ def vmlinux_proc(tree: ast.AST, module):
     with open(source_file, "r") as f:
         mod_ast = ast.parse(f.read(), filename=source_file)
 
-    for import_mod, import_node in import_statements:
-        for alias in import_node.names:
-            imported_name = alias.name
-            found = False
-            for mod_node in mod_ast.body:
-                if (
-                    isinstance(mod_node, ast.ClassDef)
-                    and mod_node.name == imported_name
-                ):
-                    process_vmlinux_class(mod_node, module, handler)
-                    found = True
-                    break
-                if isinstance(mod_node, ast.Assign):
-                    for target in mod_node.targets:
-                        if isinstance(target, ast.Name) and target.id == imported_name:
-                            process_vmlinux_assign(mod_node, module, assignments)
-                            found = True
-                            break
-                if found:
-                    break
-            if not found:
-                logger.info(
-                    f"{imported_name} not found as ClassDef or Assign in vmlinux"
-                )
+    for import_mod, import_node, imported_name, as_name in import_statements:
+        found = False
+        for mod_node in mod_ast.body:
+            if isinstance(mod_node, ast.ClassDef) and mod_node.name == imported_name:
+                process_vmlinux_class(mod_node, module, handler)
+                found = True
+                break
+            if isinstance(mod_node, ast.Assign):
+                for target in mod_node.targets:
+                    if isinstance(target, ast.Name) and target.id == imported_name:
+                        process_vmlinux_assign(mod_node, module, assignments, as_name)
+                        found = True
+                        break
+            if found:
+                break
+        if not found:
+            logger.info(f"{imported_name} not found as ClassDef or Assign in vmlinux")
 
     IRGenerator(module, handler, assignments)
     return assignments
 
 
-def process_vmlinux_assign(node, module, assignments: dict[str, AssignmentInfo]):
+def process_vmlinux_assign(
+    node, module, assignments: dict[str, AssignmentInfo], target_name=None
+):
     """Process assignments from vmlinux module."""
     # Only handle single-target assignments
     if len(node.targets) == 1 and isinstance(node.targets[0], ast.Name):
-        target_name = node.targets[0].id
+        # Use provided target_name (for aliased imports) or fall back to original name
+        if target_name is None:
+            target_name = node.targets[0].id
 
         # Handle constant value assignments
         if isinstance(node.value, ast.Constant):

pythonbpf/vmlinux_parser/ir_gen/debug_info_gen.py

@@ -21,7 +21,7 @@ def debug_info_generation(
         generated_debug_info: List of tuples (struct, debug_info) to track generated debug info
 
     Returns:
-        The generated global variable debug info
+        The generated global variable debug info, or None for unsupported types
     """
     # Set up debug info generator
     generator = DebugInfoGenerator(llvm_module)
@@ -31,23 +31,39 @@ def debug_info_generation(
         if existing_struct.name == struct.name:
             return debug_info
 
+    # Check if this is a union (not supported yet)
+    if not struct.name.startswith("struct_"):
+        logger.warning(f"Skipping debug info generation for union: {struct.name}")
+        # Create a minimal forward declaration for unions
+        union_type = generator.create_struct_type(
+            [], struct.__sizeof__() * 8, is_distinct=True
+        )
+        return union_type
+
     # Process all fields and create members for the struct
     members = []
     for field_name, field in struct.fields.items():
-        # Get appropriate debug type for this field
-        field_type = _get_field_debug_type(
-            field_name, field, generator, struct, generated_debug_info
-        )
-        # Create struct member with proper offset
-        member = generator.create_struct_member_vmlinux(
-            field_name, field_type, field.offset * 8
-        )
-        members.append(member)
+        try:
+            # Get appropriate debug type for this field
+            field_type = _get_field_debug_type(
+                field_name, field, generator, struct, generated_debug_info
+            )
 
-    if struct.name.startswith("struct_"):
-        struct_name = struct.name.removeprefix("struct_")
-    else:
-        raise ValueError("Unions are not supported in the current version")
+            # Ensure field_type is a tuple
+            if not isinstance(field_type, tuple) or len(field_type) != 2:
+                logger.error(f"Invalid field_type for {field_name}: {field_type}")
+                continue
+
+            # Create struct member with proper offset
+            member = generator.create_struct_member_vmlinux(
+                field_name, field_type, field.offset * 8
+            )
+            members.append(member)
+        except Exception as e:
+            logger.error(f"Failed to process field {field_name} in {struct.name}: {e}")
+            continue
+
+    struct_name = struct.name.removeprefix("struct_")
     # Create struct type with all members
     struct_type = generator.create_struct_type_with_name(
         struct_name, members, struct.__sizeof__() * 8, is_distinct=True
@@ -74,11 +90,17 @@ def _get_field_debug_type(
         generated_debug_info: List of already generated debug info
 
     Returns:
-        The debug info type for this field
+        A tuple of (debug_type, size_in_bits)
     """
-    # Handle complex types (arrays, pointers)
+    # Handle complex types (arrays, pointers, function pointers)
     if field.ctype_complex_type is not None:
-        if issubclass(field.ctype_complex_type, ctypes.Array):
+        # Handle function pointer types (CFUNCTYPE)
+        if callable(field.ctype_complex_type):
+            # Function pointers are represented as void pointers
+            logger.info(f"Field {field_name} is a function pointer, using void pointer")
+            void_ptr = generator.create_pointer_type(None, 64)
+            return void_ptr, 64
+        elif issubclass(field.ctype_complex_type, ctypes.Array):
             # Handle array types
             element_type, base_type_size = _get_basic_debug_type(
                 field.containing_type, generator
@@ -100,11 +122,13 @@ def _get_field_debug_type(
         for existing_struct, debug_info in generated_debug_info:
             if existing_struct.name == struct_name:
                 # Use existing debug info
-                return debug_info, existing_struct.__sizeof__()
+                return debug_info, existing_struct.__sizeof__() * 8
 
         # If not found, create a forward declaration
         # This will be completed when the actual struct is processed
-        logger.warning("Forward declaration in struct created")
+        logger.warning(
+            f"Forward declaration created for {struct_name} in {parent_struct.name}"
+        )
         forward_type = generator.create_struct_type([], 0, is_distinct=True)
         return forward_type, 0
 

pythonbpf/vmlinux_parser/ir_gen/ir_generation.py

@@ -11,6 +11,10 @@ logger = logging.getLogger(__name__)
 
 
 class IRGenerator:
+    # This field keeps track of the non_struct names to avoid duplicate name errors.
+    type_number = 0
+    unprocessed_store: list[str] = []
+
     # get the assignments dict and add this stuff to it.
     def __init__(self, llvm_module, handler: DependencyHandler, assignments):
         self.llvm_module = llvm_module
@@ -129,7 +133,19 @@ class IRGenerator:
 
         for field_name, field in struct.fields.items():
             # does not take arrays and similar types into consideration yet.
-            if field.ctype_complex_type is not None and issubclass(
+            if callable(field.ctype_complex_type):
+                # Function pointer case - generate a simple field accessor
+                field_co_re_name, returned = self._struct_name_generator(
+                    struct, field, field_index
+                )
+                field_index += 1
+                globvar = ir.GlobalVariable(
+                    self.llvm_module, ir.IntType(64), name=field_co_re_name
+                )
+                globvar.linkage = "external"
+                globvar.set_metadata("llvm.preserve.access.index", debug_info)
+                self.generated_field_names[struct.name][field_name] = globvar
+            elif field.ctype_complex_type is not None and issubclass(
                 field.ctype_complex_type, ctypes.Array
             ):
                 array_size = field.type_size
@@ -137,7 +153,7 @@ class IRGenerator:
                 if containing_type.__module__ == ctypes.__name__:
                     containing_type_size = ctypes.sizeof(containing_type)
                     if array_size == 0:
-                        field_co_re_name = self._struct_name_generator(
+                        field_co_re_name, returned = self._struct_name_generator(
                             struct, field, field_index, True, 0, containing_type_size
                         )
                         globvar = ir.GlobalVariable(
@@ -149,7 +165,7 @@ class IRGenerator:
                         field_index += 1
                         continue
                     for i in range(0, array_size):
-                        field_co_re_name = self._struct_name_generator(
+                        field_co_re_name, returned = self._struct_name_generator(
                             struct, field, field_index, True, i, containing_type_size
                         )
                         globvar = ir.GlobalVariable(
@@ -163,12 +179,28 @@ class IRGenerator:
                 array_size = field.type_size
                 containing_type = field.containing_type
                 if containing_type.__module__ == "vmlinux":
-                    containing_type_size = self.handler[
-                        containing_type.__name__
-                    ].current_offset
-                    for i in range(0, array_size):
-                        field_co_re_name = self._struct_name_generator(
-                            struct, field, field_index, True, i, containing_type_size
+                    # Unwrap all pointer layers to get the base struct type
+                    base_containing_type = containing_type
+                    while hasattr(base_containing_type, "_type_"):
+                        next_type = base_containing_type._type_
+                        # Stop if _type_ is a string (like 'c' for c_char)
+                        # TODO: stacked pointers not handl;ing ctypes check here as well
+                        if isinstance(next_type, str):
+                            break
+                        base_containing_type = next_type
+
+                    # Get the base struct name
+                    base_struct_name = (
+                        base_containing_type.__name__
+                        if hasattr(base_containing_type, "__name__")
+                        else str(base_containing_type)
+                    )
+
+                    # Look up the size using the base struct name
+                    containing_type_size = self.handler[base_struct_name].current_offset
+                    if array_size == 0:
+                        field_co_re_name, returned = self._struct_name_generator(
+                            struct, field, field_index, True, 0, containing_type_size
                         )
                         globvar = ir.GlobalVariable(
                             self.llvm_module, ir.IntType(64), name=field_co_re_name
@@ -176,9 +208,30 @@ class IRGenerator:
                         globvar.linkage = "external"
                         globvar.set_metadata("llvm.preserve.access.index", debug_info)
                         self.generated_field_names[struct.name][field_name] = globvar
-                    field_index += 1
+                        field_index += 1
+                    else:
+                        for i in range(0, array_size):
+                            field_co_re_name, returned = self._struct_name_generator(
+                                struct,
+                                field,
+                                field_index,
+                                True,
+                                i,
+                                containing_type_size,
+                            )
+                            globvar = ir.GlobalVariable(
+                                self.llvm_module, ir.IntType(64), name=field_co_re_name
+                            )
+                            globvar.linkage = "external"
+                            globvar.set_metadata(
+                                "llvm.preserve.access.index", debug_info
+                            )
+                            self.generated_field_names[struct.name][field_name] = (
+                                globvar
+                            )
+                        field_index += 1
             else:
-                field_co_re_name = self._struct_name_generator(
+                field_co_re_name, returned = self._struct_name_generator(
                     struct, field, field_index
                 )
                 field_index += 1
@@ -198,7 +251,7 @@ class IRGenerator:
         is_indexed: bool = False,
         index: int = 0,
         containing_type_size: int = 0,
-    ) -> str:
+    ) -> tuple[str, bool]:
         # TODO: Does not support Unions as well as recursive pointer and array type naming
         if is_indexed:
             name = (
@@ -208,7 +261,7 @@ class IRGenerator:
                 + "$"
                 + f"0:{field_index}:{index}"
             )
-            return name
+            return name, True
         elif struct.name.startswith("struct_"):
             name = (
                 "llvm."
@@ -217,9 +270,18 @@ class IRGenerator:
                 + "$"
                 + f"0:{field_index}"
             )
-            return name
+            return name, True
         else:
-            print(self.handler[struct.name])
-            raise TypeError(
-                "Name generation cannot occur due to type name not starting with struct"
+            logger.warning(
+                "Blindly handling non-struct type to avoid type errors in vmlinux IR generation. Possibly a union."
             )
+            self.type_number += 1
+            unprocessed_type = "unprocessed_type_" + str(self.handler[struct.name].name)
+            if self.unprocessed_store.__contains__(unprocessed_type):
+                return unprocessed_type + "_" + str(self.type_number), False
+            else:
+                self.unprocessed_store.append(unprocessed_type)
+                return unprocessed_type, False
+            # raise TypeError(
+            #     "Name generation cannot occur due to type name not starting with struct"
+            # )

tests/c-form/Makefile

@@ -3,21 +3,20 @@ CFLAGS := -emit-llvm -target bpf -c
 
 SRC := $(wildcard *.bpf.c)
 LL := $(SRC:.bpf.c=.bpf.ll)
-LL2 := $(SRC:.bpf.c=.bpf.o2.ll)
 OBJ := $(SRC:.bpf.c=.bpf.o)
-
+LL0 := $(SRC:.bpf.c=.bpf.o0.ll)
 .PHONY: all clean
 
-all: $(LL) $(OBJ) $(LL2)
+all: $(LL) $(OBJ) $(LL0)
 
 %.bpf.o: %.bpf.c
 	$(BPF_CLANG) -O2 -g -target bpf -c $< -o $@
 
 %.bpf.ll: %.bpf.c
-	$(BPF_CLANG) -O0 $(CFLAGS) -g -S $< -o $@
+	$(BPF_CLANG) $(CFLAGS) -O2 -g -S $< -o $@
 
-%.bpf.o2.ll: %.bpf.c
-	$(BPF_CLANG) -O2 $(CFLAGS) -g -S $< -o $@
+%.bpf.o0.ll: %.bpf.c
+	$(BPF_CLANG) $(CFLAGS) -O0 -g -S $< -o $@
 
 clean:
-	rm -f $(LL) $(OBJ) $(LL2)
+	rm -f $(LL) $(OBJ) $(LL0)

tests/c-form/requests.bpf.c

@@ -0,0 +1,15 @@
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+char LICENSE[] SEC("license") = "GPL";
+
+SEC("kprobe/blk_mq_start_request")
+int example(struct pt_regs *ctx)
+{
+    struct request *req = (struct request *)(ctx->di);
+    u32 data_len = req->__data_len;
+    bpf_printk("data length %u\n", data_len);
+
+    return 0;
+}

tests/failing_tests/vmlinux/assignment_handling.py

@@ -0,0 +1,22 @@
+from vmlinux import XDP_PASS
+from pythonbpf import bpf, section, bpfglobal, compile_to_ir
+import logging
+from ctypes import c_int64, c_void_p
+
+
+@bpf
+@section("kprobe/blk_mq_start_request")
+def example(ctx: c_void_p) -> c_int64:
+    d = XDP_PASS  # This gives an error, but
+    e = XDP_PASS + 0  # this does not
+    print(f"test1 {e} test2 {d}")
+    return c_int64(0)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile_to_ir("assignment_handling.py", "assignment_handling.ll", loglevel=logging.INFO)

tests/failing_tests/vmlinux/requests.py

@@ -0,0 +1,22 @@
+from vmlinux import struct_request, struct_pt_regs
+from pythonbpf import bpf, section, bpfglobal, compile_to_ir
+import logging
+from ctypes import c_int64
+
+
+@bpf
+@section("kprobe/blk_mq_start_request")
+def example(ctx: struct_pt_regs) -> c_int64:
+    req = struct_request(ctx.di)
+    c = req.__data_len
+    print(f"data length {c}")
+    return c_int64(0)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile_to_ir("requests.py", "requests.ll", loglevel=logging.INFO)

tests/passing_tests/hash_map_struct.py

@@ -0,0 +1,42 @@
+from pythonbpf import bpf, section, struct, bpfglobal, compile, map
+from pythonbpf.maps import HashMap
+from pythonbpf.helper import pid
+from ctypes import c_void_p, c_int64
+
+
+@bpf
+@struct
+class val_type:
+    counter: c_int64
+    shizzle: c_int64
+
+
+@bpf
+@map
+def last() -> HashMap:
+    return HashMap(key=val_type, value=c_int64, max_entries=16)
+
+
+@bpf
+@section("tracepoint/syscalls/sys_enter_clone")
+def hello_world(ctx: c_void_p) -> c_int64:
+    obj = val_type()
+    obj.counter, obj.shizzle = 42, 96
+    t = last.lookup(obj)
+    if t:
+        print(f"Found existing entry: counter={obj.counter}, pid={t}")
+        last.delete(obj)
+        return 0  # type: ignore [return-value]
+    val = pid()
+    last.update(obj, val)
+    print(f"Map updated!, {obj.counter}, {obj.shizzle}, {val}")
+    return 0  # type: ignore [return-value]
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()

tests/passing_tests/vmlinux/requests2.py

@@ -0,0 +1,21 @@
+from vmlinux import struct_pt_regs
+from pythonbpf import bpf, section, bpfglobal, compile_to_ir
+import logging
+from ctypes import c_int64
+
+
+@bpf
+@section("kprobe/blk_mq_start_request")
+def example(ctx: struct_pt_regs) -> c_int64:
+    req = ctx.di
+    print(f"data length {req}")
+    return c_int64(0)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile_to_ir("requests2.py", "requests2.ll", loglevel=logging.INFO)