Pass down structs_sym_tab to maps_debug_info, allow vmlinux enums to be used in an indexed format for map declaration

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "pythonbpf"
-version = "0.1.7"
+version = "0.1.6"
 description = "Reduced Python frontend for eBPF"
 authors = [
   { name = "r41k0u", email="pragyanshchaturvedi18@gmail.com" },
@@ -29,7 +29,7 @@ license = {text = "Apache-2.0"}
 requires-python = ">=3.10"
 
 dependencies = [
-  "llvmlite>=0.45",
+  "llvmlite",
   "astpretty",
   "pylibbpf"
 ]

pythonbpf/allocation_pass.py

@@ -7,7 +7,6 @@ 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__)
 
@@ -26,9 +25,7 @@ def create_targets_and_rvals(stmt):
     return stmt.targets, [stmt.value]
 
 
-def handle_assign_allocation(
+def handle_assign_allocation(builder, stmt, local_sym_tab, structs_sym_tab):
-    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)}")
@@ -58,9 +55,7 @@ def handle_assign_allocation(
 
         # Determine type and allocate based on rval
         if isinstance(rval, ast.Call):
-            _allocate_for_call(
+            _allocate_for_call(builder, var_name, rval, local_sym_tab, structs_sym_tab)
-                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):
@@ -79,9 +74,7 @@ def handle_assign_allocation(
             )
 
 
-def _allocate_for_call(
+def _allocate_for_call(builder, var_name, rval, local_sym_tab, structs_sym_tab):
-    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):
@@ -118,91 +111,20 @@ def _allocate_for_call(
             local_sym_tab[var_name] = LocalSymbol(var, struct_info.ir_type, call_type)
             logger.info(f"Pre-allocated {var_name} for struct {call_type}")
 
-        elif VmlinuxHandlerRegistry.is_vmlinux_struct(call_type):
-            # When calling struct_name(pointer), we're doing a cast, not construction
-            # So we allocate as a pointer (i64) not as the actual struct
-            var = builder.alloca(ir.IntType(64), name=var_name)
-            var.align = 8
-            local_sym_tab[var_name] = LocalSymbol(
-                var, ir.IntType(64), VmlinuxHandlerRegistry.get_struct_type(call_type)
-            )
-            logger.info(
-                f"Pre-allocated {var_name} for vmlinux struct pointer cast to {call_type}"
-            )
-
         else:
             logger.warning(f"Unknown call type for allocation: {call_type}")
 
     elif isinstance(rval.func, ast.Attribute):
         # Map method calls - need double allocation for ptr handling
-        _allocate_for_map_method(
+        _allocate_for_map_method(builder, var_name, local_sym_tab)
-            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(
+def _allocate_for_map_method(builder, var_name, local_sym_tab):
-    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, value_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)
@@ -213,9 +135,7 @@ def _allocate_for_map_method_fallback(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(
+    logger.info(f"Pre-allocated {var_name} and {var_name}_tmp for map method")
-        f"Pre-allocated {var_name} and {var_name}_tmp for map method (fallback)"
-    )
 
 
 def _allocate_for_constant(builder, var_name, rval, local_sym_tab):
@@ -337,6 +257,13 @@ def _allocate_for_attribute(builder, var_name, rval, local_sym_tab, structs_sym_
                 VmlinuxHandlerRegistry.get_field_type(vmlinux_struct_name, field_name)
             )
             field_ir, field = field_type
+            # TODO: For now, we only support integer type allocations.
+            # This always assumes first argument of function to be the context struct
+            base_ptr = builder.function.args[0]
+            local_sym_tab[
+                struct_var
+            ].var = base_ptr  # This is repurposing of var to store the pointer of the base type
+            local_sym_tab[struct_var].ir_type = field_ir
 
             # Determine the actual IR type based on the field's type
             actual_ir_type = None
@@ -391,14 +318,12 @@ def _allocate_for_attribute(builder, var_name, rval, local_sym_tab, structs_sym_
                 )
                 actual_ir_type = ir.IntType(64)
 
-            # Allocate with the actual IR type
+            # Allocate with the actual IR type, not the GlobalVariable
             var = _allocate_with_type(builder, var_name, actual_ir_type)
-            local_sym_tab[var_name] = LocalSymbol(
+            local_sym_tab[var_name] = LocalSymbol(var, actual_ir_type, field)
-                var, actual_ir_type, field
-            )  # <-- Store Field metadata
 
             logger.info(
-                f"Pre-allocated {var_name} as {actual_ir_type} from vmlinux struct {vmlinux_struct_name}.{field_name}"
+                f"Pre-allocated {var_name} from vmlinux struct {vmlinux_struct_name}.{field_name}"
             )
             return
         else:

pythonbpf/assign_pass.py

@@ -1,7 +1,5 @@
 import ast
 import logging
-from inspect import isclass
-
 from llvmlite import ir
 from pythonbpf.expr import eval_expr
 from pythonbpf.helper import emit_probe_read_kernel_str_call
@@ -150,30 +148,8 @@ def handle_variable_assignment(
         return False
 
     val, val_type = val_result
-    logger.info(
+    logger.info(f"Evaluated value for {var_name}: {val} of type {val_type}, {var_type}")
-        f"Evaluated value for {var_name}: {val} of type {val_type}, expected {var_type}"
-    )
-
     if val_type != var_type:
-        # Handle vmlinux struct pointers - they're represented as Python classes but are i64 pointers
-        if isclass(val_type) and (val_type.__module__ == "vmlinux"):
-            logger.info("Handling vmlinux struct pointer assignment")
-            # vmlinux struct pointers: val is a pointer, need to convert to i64
-            if isinstance(var_type, ir.IntType) and var_type.width == 64:
-                # Convert pointer to i64 using ptrtoint
-                if isinstance(val.type, ir.PointerType):
-                    val = builder.ptrtoint(val, ir.IntType(64))
-                    logger.info(
-                        "Converted vmlinux struct pointer to i64 using ptrtoint"
-                    )
-                builder.store(val, var_ptr)
-                logger.info(f"Assigned vmlinux struct pointer to {var_name} (i64)")
-                return True
-            else:
-                logger.error(
-                    f"Type mismatch: vmlinux struct pointer requires i64, got {var_type}"
-                )
-                return False
         if isinstance(val_type, Field):
             logger.info("Handling assignment to struct field")
             # Special handling for struct_xdp_md i32 fields that are zero-extended to i64

pythonbpf/codegen.py

@@ -25,7 +25,7 @@ import re
 
 logger: Logger = logging.getLogger(__name__)
 
-VERSION = "v0.1.7"
+VERSION = "v0.1.6"
 
 
 def finalize_module(original_str):

pythonbpf/debuginfo/debug_info_generator.py

@@ -49,10 +49,6 @@ 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/expr/expr_pass.py

@@ -12,8 +12,8 @@ from .type_normalization import (
     get_base_type_and_depth,
     deref_to_depth,
 )
+from pythonbpf.vmlinux_parser.assignment_info import Field
 from .vmlinux_registry import VmlinuxHandlerRegistry
-from ..vmlinux_parser.dependency_node import Field
 
 logger: Logger = logging.getLogger(__name__)
 
@@ -61,7 +61,6 @@ def _handle_constant_expr(module, builder, expr: ast.Constant):
 
 
 def _handle_attribute_expr(
-    func,
     expr: ast.Attribute,
     local_sym_tab: Dict,
     structs_sym_tab: Dict,
@@ -77,89 +76,6 @@ def _handle_attribute_expr(
             logger.info(
                 f"Variable type: {var_type}, Variable ptr: {var_ptr}, Variable Metadata: {var_metadata}"
             )
-            # Check if this is a pointer to a struct (from map lookup)
-            if (
-                isinstance(var_type, ir.PointerType)
-                and var_metadata
-                and isinstance(var_metadata, str)
-            ):
-                if var_metadata in structs_sym_tab:
-                    logger.info(
-                        f"Handling pointer to struct {var_metadata} from map lookup"
-                    )
-
-                    if func is None:
-                        raise ValueError(
-                            f"func parameter required for null-safe pointer access to {var_name}.{attr_name}"
-                        )
-
-                    # Load the pointer value (ptr<struct>)
-                    struct_ptr = builder.load(var_ptr)
-
-                    # Create blocks for null check
-                    null_check_block = builder.block
-                    not_null_block = func.append_basic_block(
-                        name=f"{var_name}_not_null"
-                    )
-                    merge_block = func.append_basic_block(name=f"{var_name}_merge")
-
-                    # Check if pointer is null
-                    null_ptr = ir.Constant(struct_ptr.type, None)
-                    is_not_null = builder.icmp_signed("!=", struct_ptr, null_ptr)
-                    logger.info(f"Inserted null check for pointer {var_name}")
-
-                    builder.cbranch(is_not_null, not_null_block, merge_block)
-
-                    # Not-null block: Access the field
-                    builder.position_at_end(not_null_block)
-
-                    # Get struct metadata
-                    metadata = structs_sym_tab[var_metadata]
-                    struct_ptr = builder.bitcast(
-                        struct_ptr, metadata.ir_type.as_pointer()
-                    )
-
-                    if attr_name not in metadata.fields:
-                        raise ValueError(
-                            f"Field '{attr_name}' not found in struct '{var_metadata}'"
-                        )
-
-                    # GEP to field
-                    field_gep = metadata.gep(builder, struct_ptr, attr_name)
-
-                    # Load field value
-                    field_val = builder.load(field_gep)
-                    field_type = metadata.field_type(attr_name)
-
-                    logger.info(
-                        f"Loaded field {attr_name} from struct pointer, type: {field_type}"
-                    )
-
-                    # Branch to merge
-                    not_null_after_load = builder.block
-                    builder.branch(merge_block)
-
-                    # Merge block: PHI node for the result
-                    builder.position_at_end(merge_block)
-                    phi = builder.phi(field_type, name=f"{var_name}_{attr_name}")
-
-                    # If null, return zero/default value
-                    if isinstance(field_type, ir.IntType):
-                        zero_value = ir.Constant(field_type, 0)
-                    elif isinstance(field_type, ir.PointerType):
-                        zero_value = ir.Constant(field_type, None)
-                    elif isinstance(field_type, ir.ArrayType):
-                        # For arrays, we can't easily create a zero constant
-                        # This case is tricky - for now, just use undef
-                        zero_value = ir.Constant(field_type, ir.Undefined)
-                    else:
-                        zero_value = ir.Constant(field_type, ir.Undefined)
-
-                    phi.add_incoming(zero_value, null_check_block)
-                    phi.add_incoming(field_val, not_null_after_load)
-
-                    logger.info(f"Created PHI node for {var_name}.{attr_name}")
-                    return phi, field_type
             if (
                 hasattr(var_metadata, "__module__")
                 and var_metadata.__module__ == "vmlinux"
@@ -173,16 +89,8 @@ def _handle_attribute_expr(
                     return vmlinux_result
                 else:
                     raise RuntimeError("Vmlinux struct did not process successfully")
-
+            metadata = structs_sym_tab[var_metadata]
-            elif isinstance(var_metadata, Field):
+            if attr_name in metadata.fields:
-                logger.error(
-                    f"Cannot access field '{attr_name}' on already-loaded field value '{var_name}'"
-                )
-                return None
-
-            # Regular user-defined struct
-            metadata = structs_sym_tab.get(var_metadata)
-            if metadata and attr_name in metadata.fields:
                 gep = metadata.gep(builder, var_ptr, attr_name)
                 val = builder.load(gep)
                 field_type = metadata.field_type(attr_name)
@@ -617,66 +525,6 @@ def _handle_boolean_op(
         return None
 
 
-# ============================================================================
-# VMLinux casting
-# ============================================================================
-
-
-def _handle_vmlinux_cast(
-    func,
-    module,
-    builder,
-    expr,
-    local_sym_tab,
-    map_sym_tab,
-    structs_sym_tab=None,
-):
-    # handle expressions such as struct_request(ctx.di) where struct_request is a vmlinux
-    # struct and ctx.di is a pointer to a struct but is actually represented as a c_uint64
-    # which needs to be cast to a pointer. This is also a field of another vmlinux struct
-    """Handle vmlinux struct cast expressions like struct_request(ctx.di)."""
-    if len(expr.args) != 1:
-        logger.info("vmlinux struct cast takes exactly one argument")
-        return None
-
-    # Get the struct name
-    struct_name = expr.func.id
-
-    # Evaluate the argument (e.g., ctx.di which is a c_uint64)
-    arg_result = eval_expr(
-        func,
-        module,
-        builder,
-        expr.args[0],
-        local_sym_tab,
-        map_sym_tab,
-        structs_sym_tab,
-    )
-
-    if arg_result is None:
-        logger.info("Failed to evaluate argument to vmlinux struct cast")
-        return None
-
-    arg_val, arg_type = arg_result
-    # Get the vmlinux struct type
-    vmlinux_struct_type = VmlinuxHandlerRegistry.get_struct_type(struct_name)
-    if vmlinux_struct_type is None:
-        logger.error(f"Failed to get vmlinux struct type for {struct_name}")
-        return None
-    # Cast the integer/value to a pointer to the struct
-    # If arg_val is an integer type, we need to inttoptr it
-    ptr_type = ir.PointerType()
-    # TODO: add a integer check here later
-    if ctypes_to_ir(arg_type.type.__name__):
-        # Cast integer to pointer
-        casted_ptr = builder.inttoptr(arg_val, ptr_type)
-    else:
-        logger.error(f"Unsupported type for vmlinux cast: {arg_type}")
-        return None
-
-    return casted_ptr, vmlinux_struct_type
-
-
 # ============================================================================
 # Expression Dispatcher
 # ============================================================================
@@ -697,18 +545,6 @@ def eval_expr(
     elif isinstance(expr, ast.Constant):
         return _handle_constant_expr(module, builder, expr)
     elif isinstance(expr, ast.Call):
-        if isinstance(expr.func, ast.Name) and VmlinuxHandlerRegistry.is_vmlinux_struct(
-            expr.func.id
-        ):
-            return _handle_vmlinux_cast(
-                func,
-                module,
-                builder,
-                expr,
-                local_sym_tab,
-                map_sym_tab,
-                structs_sym_tab,
-            )
         if isinstance(expr.func, ast.Name) and expr.func.id == "deref":
             return _handle_deref_call(expr, local_sym_tab, builder)
 
@@ -732,9 +568,7 @@ def eval_expr(
         logger.warning(f"Unknown call: {ast.dump(expr)}")
         return None
     elif isinstance(expr, ast.Attribute):
-        return _handle_attribute_expr(
+        return _handle_attribute_expr(expr, local_sym_tab, structs_sym_tab, builder)
-            func, expr, local_sym_tab, structs_sym_tab, builder
-        )
     elif isinstance(expr, ast.BinOp):
         return _handle_binary_op(
             func,

pythonbpf/expr/ir_ops.py

@@ -48,10 +48,3 @@ def deref_to_depth(func, builder, val, target_depth):
         cur_val = phi
         cur_type = pointee_type
     return cur_val
-
-
-def deref_struct_ptr(
-    func, builder, struct_ptr, struct_metadata, field_name, structs_sym_tab
-):
-    """Dereference a pointer to a struct type."""
-    return deref_to_depth(func, builder, struct_ptr, 1)

pythonbpf/functions/functions_pass.py

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

pythonbpf/helper/__init__.py

@@ -1,10 +1,6 @@
 from .helper_registry import HelperHandlerRegistry
 from .helper_utils import reset_scratch_pool
-from .bpf_helper_handler import (
+from .bpf_helper_handler import handle_helper_call, emit_probe_read_kernel_str_call
-    handle_helper_call,
-    emit_probe_read_kernel_str_call,
-    emit_probe_read_kernel_call,
-)
 from .helpers import (
     ktime,
     pid,
@@ -78,7 +74,6 @@ __all__ = [
     "reset_scratch_pool",
     "handle_helper_call",
     "emit_probe_read_kernel_str_call",
-    "emit_probe_read_kernel_call",
     "ktime",
     "pid",
     "deref",

pythonbpf/helper/bpf_helper_handler.py

@@ -34,7 +34,6 @@ class BPFHelperID(Enum):
     BPF_PERF_EVENT_OUTPUT = 25
     BPF_GET_STACK = 67
     BPF_PROBE_READ_KERNEL_STR = 115
-    BPF_PROBE_READ_KERNEL = 113
     BPF_RINGBUF_OUTPUT = 130
     BPF_RINGBUF_RESERVE = 131
     BPF_RINGBUF_SUBMIT = 132
@@ -575,75 +574,6 @@ def bpf_probe_read_kernel_str_emitter(
     return result, ir.IntType(64)
 
 
-def emit_probe_read_kernel_call(builder, dst_ptr, dst_size, src_ptr):
-    """Emit LLVM IR call to bpf_probe_read_kernel"""
-
-    fn_type = ir.FunctionType(
-        ir.IntType(64),
-        [ir.PointerType(), ir.IntType(32), ir.PointerType()],
-        var_arg=False,
-    )
-    fn_ptr = builder.inttoptr(
-        ir.Constant(ir.IntType(64), BPFHelperID.BPF_PROBE_READ_KERNEL.value),
-        ir.PointerType(fn_type),
-    )
-
-    result = builder.call(
-        fn_ptr,
-        [
-            builder.bitcast(dst_ptr, ir.PointerType()),
-            ir.Constant(ir.IntType(32), dst_size),
-            builder.bitcast(src_ptr, ir.PointerType()),
-        ],
-        tail=False,
-    )
-
-    logger.info(f"Emitted bpf_probe_read_kernel (size={dst_size})")
-    return result
-
-
-@HelperHandlerRegistry.register(
-    "probe_read_kernel",
-    param_types=[
-        ir.PointerType(ir.IntType(8)),
-        ir.PointerType(ir.IntType(8)),
-    ],
-    return_type=ir.IntType(64),
-)
-def bpf_probe_read_kernel_emitter(
-    call,
-    map_ptr,
-    module,
-    builder,
-    func,
-    local_sym_tab=None,
-    struct_sym_tab=None,
-    map_sym_tab=None,
-):
-    """Emit LLVM IR for bpf_probe_read_kernel helper."""
-
-    if len(call.args) != 2:
-        raise ValueError(
-            f"probe_read_kernel expects 2 args (dst, src), got {len(call.args)}"
-        )
-
-    # Get destination buffer (char array -> i8*)
-    dst_ptr, dst_size = get_or_create_ptr_from_arg(
-        func, module, call.args[0], builder, local_sym_tab, map_sym_tab, struct_sym_tab
-    )
-
-    # Get source pointer (evaluate expression)
-    src_ptr, src_type = get_ptr_from_arg(
-        call.args[1], func, module, builder, local_sym_tab, map_sym_tab, struct_sym_tab
-    )
-
-    # Emit the helper call
-    result = emit_probe_read_kernel_call(builder, dst_ptr, dst_size, src_ptr)
-
-    logger.info(f"Emitted bpf_probe_read_kernel (size={dst_size})")
-    return result, ir.IntType(64)
-
-
 @HelperHandlerRegistry.register(
     "random",
     param_types=[],

pythonbpf/maps/map_debug_info.py

@@ -1,31 +1,22 @@
-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
+        array_type if "key_size" in map_params else ulong_type, 64
-        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
+        array_type if "value_size" in map_params else ulong_type, 64
-        if "value_size" in map_params
-        else _get_key_val_dbg_type(map_params.get("value"), generator, structs_sym_tab),
-        64,
     )
 
     elements_arr = []
@@ -106,66 +97,3 @@ 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:
-        logger.info(f"Found struct named {name}, generating debug type")
-        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 * 8, 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, params=map_params)
+    return MapSymbol(type=map_params["type"], sym=map_global)
 
 
 def _parse_map_params(rval, expected_args=None):
@@ -105,9 +105,7 @@ 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(
+    create_ringbuf_debug_info(module, map_global.sym, map_name, map_params)
-        module, map_global.sym, map_name, map_params, structs_sym_tab
-    )
     return map_global
 
 
@@ -121,7 +119,7 @@ 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, structs_sym_tab)
+    create_map_debug_info(module, map_global.sym, map_name, map_params)
     return map_global
 
 

pythonbpf/maps/maps_utils.py

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

pythonbpf/vmlinux_parser/class_handler.py

@@ -16,33 +16,6 @@ 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,
@@ -185,90 +158,13 @@ def process_vmlinux_post_ast(
                         if hasattr(elem_type, "_length_") and is_complex_type:
                             type_length = elem_type._length_
 
-                        # Unwrap all pointer layers to get the base type for dependency tracking
+                        if containing_type.__module__ == "vmlinux":
-                        base_type = unwrap_pointer_type(elem_type)
+                            new_dep_node.add_dependent(
-                        base_type_module = getattr(base_type, "__module__", None)
+                                elem_type._type_.__name__
-
+                                if hasattr(elem_type._type_, "__name__")
-                        if base_type_module == "vmlinux":
+                                else str(elem_type._type_)
-                            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
@@ -280,20 +176,57 @@ def process_vmlinux_post_ast(
                                     )
                             else:
                                 raise TypeError("Unsupported ctypes subclass")
-
-                            # 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 ImportError(
-                                f"Unsupported module of {base_type}: {base_type_module}"
+                                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}")
+                            new_dep_node.set_field_ready(elem_name, True)
+                        else:
+                            raise TypeError(
+                                "Module not supported in recursive resolution"
                             )
                     else:
                         new_dep_node.add_dependent(
@@ -312,12 +245,9 @@ 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(f"UNSUPPORTED Module {module_name}")
+        raise ImportError("UNSUPPORTED Module")
 
     logger.info(
         f"{current_symbol_name} processed and handler readiness {handler.is_ready}"

pythonbpf/vmlinux_parser/import_detector.py

@@ -11,9 +11,7 @@ from .class_handler import process_vmlinux_class
 logger = logging.getLogger(__name__)
 
 
-def detect_import_statement(
+def detect_import_statement(tree: ast.AST) -> list[tuple[str, ast.ImportFrom]]:
-    tree: ast.AST,
-) -> list[tuple[str, ast.ImportFrom, str, str]]:
     """
     Parse AST and detect import statements from vmlinux.
 
@@ -27,7 +25,7 @@ def detect_import_statement(
         List of tuples containing (module_name, imported_item) for each vmlinux import
 
     Raises:
-        SyntaxError: If import * is used
+        SyntaxError: If multiple imports from vmlinux are attempted or import * is used
     """
     vmlinux_imports = []
 
@@ -42,19 +40,28 @@ def detect_import_statement(
                         "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."
                     )
 
-                # Support multiple imports: from vmlinux import A, B, C
+                # Valid single import
                 for alias in node.names:
                     import_name = alias.name
-                    # Use alias if provided, otherwise use the original name
+                    # Use alias if provided, otherwise use the original name (commented)
-                    as_name = alias.asname if alias.asname else alias.name
+                    # as_name = alias.asname if alias.asname else alias.name
-                    vmlinux_imports.append(("vmlinux", node, import_name, as_name))
+                    vmlinux_imports.append(("vmlinux", node))
-                    logger.info(f"Found vmlinux import: {import_name} as {as_name}")
+                    logger.info(f"Found vmlinux import: {import_name}")
 
         # Handle "import vmlinux" statements (not typical but should be rejected)
         elif isinstance(node, ast.Import):
@@ -96,37 +103,40 @@ 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, imported_name, as_name in import_statements:
+    for import_mod, import_node in import_statements:
-        found = False
+        for alias in import_node.names:
-        for mod_node in mod_ast.body:
+            imported_name = alias.name
-            if isinstance(mod_node, ast.ClassDef) and mod_node.name == imported_name:
+            found = False
-                process_vmlinux_class(mod_node, module, handler)
+            for mod_node in mod_ast.body:
-                found = True
+                if (
-                break
+                    isinstance(mod_node, ast.ClassDef)
-            if isinstance(mod_node, ast.Assign):
+                    and mod_node.name == imported_name
-                for target in mod_node.targets:
+                ):
-                    if isinstance(target, ast.Name) and target.id == imported_name:
+                    process_vmlinux_class(mod_node, module, handler)
-                        process_vmlinux_assign(mod_node, module, assignments, as_name)
+                    found = True
-                        found = True
+                    break
-                        break
+                if isinstance(mod_node, ast.Assign):
-            if found:
+                    for target in mod_node.targets:
-                break
+                        if isinstance(target, ast.Name) and target.id == imported_name:
-        if not found:
+                            process_vmlinux_assign(mod_node, module, assignments)
-            logger.info(f"{imported_name} not found as ClassDef or Assign in vmlinux")
+                            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(
+def process_vmlinux_assign(node, module, assignments: dict[str, AssignmentInfo]):
-    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):
-        # Use provided target_name (for aliased imports) or fall back to original name
+        target_name = node.targets[0].id
-        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, or None for unsupported types
+        The generated global variable debug info
     """
     # Set up debug info generator
     generator = DebugInfoGenerator(llvm_module)
@@ -31,42 +31,23 @@ 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)
 
-    sorted_fields = sorted(struct.fields.items(), key=lambda item: item[1].offset)
+    if struct.name.startswith("struct_"):
-
+        struct_name = struct.name.removeprefix("struct_")
-    for field_name, field in sorted_fields:
+    else:
-        try:
+        raise ValueError("Unions are not supported in the current version")
-            # Get appropriate debug type for this field
-            field_type = _get_field_debug_type(
-                field_name, field, generator, struct, generated_debug_info
-            )
-
-            # 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
@@ -93,19 +74,11 @@ def _get_field_debug_type(
         generated_debug_info: List of already generated debug info
 
     Returns:
-        A tuple of (debug_type, size_in_bits)
+        The debug info type for this field
     """
-    # Handle complex types (arrays, pointers, function pointers)
+    # Handle complex types (arrays, pointers)
     if field.ctype_complex_type is not None:
-        # Handle function pointer types (CFUNCTYPE)
+        if issubclass(field.ctype_complex_type, ctypes.Array):
-        if callable(field.ctype_complex_type):
-            # Function pointers are represented as void pointers
-            logger.warning(
-                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
@@ -127,13 +100,11 @@ 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__() * 8
+                return debug_info, existing_struct.__sizeof__()
 
         # If not found, create a forward declaration
         # This will be completed when the actual struct is processed
-        logger.info(
+        logger.warning("Forward declaration in struct created")
-            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

@@ -11,10 +11,6 @@ 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
@@ -133,19 +129,7 @@ class IRGenerator:
 
         for field_name, field in struct.fields.items():
             # does not take arrays and similar types into consideration yet.
-            if callable(field.ctype_complex_type):
+            if field.ctype_complex_type is not None and issubclass(
-                # 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
@@ -153,7 +137,7 @@ class IRGenerator:
                 if containing_type.__module__ == ctypes.__name__:
                     containing_type_size = ctypes.sizeof(containing_type)
                     if array_size == 0:
-                        field_co_re_name, returned = self._struct_name_generator(
+                        field_co_re_name = self._struct_name_generator(
                             struct, field, field_index, True, 0, containing_type_size
                         )
                         globvar = ir.GlobalVariable(
@@ -165,7 +149,7 @@ class IRGenerator:
                         field_index += 1
                         continue
                     for i in range(0, array_size):
-                        field_co_re_name, returned = self._struct_name_generator(
+                        field_co_re_name = self._struct_name_generator(
                             struct, field, field_index, True, i, containing_type_size
                         )
                         globvar = ir.GlobalVariable(
@@ -179,28 +163,12 @@ class IRGenerator:
                 array_size = field.type_size
                 containing_type = field.containing_type
                 if containing_type.__module__ == "vmlinux":
-                    # Unwrap all pointer layers to get the base struct type
+                    containing_type_size = self.handler[
-                    base_containing_type = containing_type
+                        containing_type.__name__
-                    while hasattr(base_containing_type, "_type_"):
+                    ].current_offset
-                        next_type = base_containing_type._type_
+                    for i in range(0, array_size):
-                        # Stop if _type_ is a string (like 'c' for c_char)
+                        field_co_re_name = self._struct_name_generator(
-                        # TODO: stacked pointers not handl;ing ctypes check here as well
+                            struct, field, field_index, True, i, containing_type_size
-                        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
@@ -208,30 +176,9 @@ 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, returned = self._struct_name_generator(
+                field_co_re_name = self._struct_name_generator(
                     struct, field, field_index
                 )
                 field_index += 1
@@ -251,7 +198,7 @@ class IRGenerator:
         is_indexed: bool = False,
         index: int = 0,
         containing_type_size: int = 0,
-    ) -> tuple[str, bool]:
+    ) -> str:
         # TODO: Does not support Unions as well as recursive pointer and array type naming
         if is_indexed:
             name = (
@@ -261,7 +208,7 @@ class IRGenerator:
                 + "$"
                 + f"0:{field_index}:{index}"
             )
-            return name, True
+            return name
         elif struct.name.startswith("struct_"):
             name = (
                 "llvm."
@@ -270,18 +217,9 @@ class IRGenerator:
                 + "$"
                 + f"0:{field_index}"
             )
-            return name, True
+            return name
         else:
-            logger.warning(
+            print(self.handler[struct.name])
-                "Blindly handling non-struct type to avoid type errors in vmlinux IR generation. Possibly a union."
+            raise TypeError(
+                "Name generation cannot occur due to type name not starting with struct"
             )
-            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"
-            # )

pythonbpf/vmlinux_parser/vmlinux_exports_handler.py

@@ -94,140 +94,17 @@ class VmlinuxHandler:
                 f"Attempting to access field {field_name} of possible vmlinux struct {struct_var_name}"
             )
             python_type: type = var_info.metadata
-            # Check if this is a context field (ctx) or a cast struct
+            struct_name = python_type.__name__
-            is_context_field = var_info.var is None
+            globvar_ir, field_data = self.get_field_type(struct_name, field_name)
-
+            builder.function.args[0].type = ir.PointerType(ir.IntType(8))
-            if is_context_field:
+            field_ptr = self.load_ctx_field(
-                # Handle context field access (original behavior)
+                builder, builder.function.args[0], globvar_ir, field_data, struct_name
-                struct_name = python_type.__name__
+            )
-                globvar_ir, field_data = self.get_field_type(struct_name, field_name)
+            # Return pointer to field and field type
-                builder.function.args[0].type = ir.PointerType(ir.IntType(8))
+            return field_ptr, field_data
-                field_ptr = self.load_ctx_field(
-                    builder,
-                    builder.function.args[0],
-                    globvar_ir,
-                    field_data,
-                    struct_name,
-                )
-                return field_ptr, field_data
-            else:
-                # Handle cast struct field access
-                struct_name = python_type.__name__
-                globvar_ir, field_data = self.get_field_type(struct_name, field_name)
-
-                # Handle cast struct field access (use bpf_probe_read_kernel)
-                # Load the struct pointer from the local variable
-                struct_ptr = builder.load(var_info.var)
-
-                # Use bpf_probe_read_kernel for non-context struct field access
-                field_value = self.load_struct_field(
-                    builder, struct_ptr, globvar_ir, field_data, struct_name
-                )
-                # Return field value and field type
-                return field_value, field_data
         else:
             raise RuntimeError("Variable accessed not found in symbol table")
 
-    @staticmethod
-    def load_struct_field(
-        builder, struct_ptr_int, offset_global, field_data, struct_name=None
-    ):
-        """
-        Generate LLVM IR to load a field from a regular (non-context) struct using bpf_probe_read_kernel.
-
-        Args:
-            builder: llvmlite IRBuilder instance
-            struct_ptr_int: The struct pointer as an i64 value (already loaded from alloca)
-            offset_global: Global variable containing the field offset (i64)
-            field_data: contains data about the field
-            struct_name: Name of the struct being accessed (optional)
-        Returns:
-            The loaded value
-        """
-
-        # Load the offset value
-        offset = builder.load(offset_global)
-
-        # Convert i64 to pointer type (BPF stores pointers as i64)
-        i8_ptr_type = ir.PointerType(ir.IntType(8))
-        struct_ptr = builder.inttoptr(struct_ptr_int, i8_ptr_type)
-
-        # GEP with offset to get field pointer
-        field_ptr = builder.gep(
-            struct_ptr,
-            [offset],
-            inbounds=False,
-        )
-
-        # Determine the appropriate field size based on field information
-        field_size_bytes = 8  # Default to 8 bytes (64-bit)
-        int_width = 64  # Default to 64-bit
-        needs_zext = False
-
-        if field_data is not None:
-            # Try to determine the size from field metadata
-            if field_data.type.__module__ == ctypes.__name__:
-                try:
-                    field_size_bytes = ctypes.sizeof(field_data.type)
-                    field_size_bits = field_size_bytes * 8
-
-                    if field_size_bits in [8, 16, 32, 64]:
-                        int_width = field_size_bits
-                        logger.info(
-                            f"Determined field size: {int_width} bits ({field_size_bytes} bytes)"
-                        )
-
-                        # Special handling for struct_xdp_md i32 fields
-                        if struct_name == "struct_xdp_md" and int_width == 32:
-                            needs_zext = True
-                            logger.info(
-                                "struct_xdp_md i32 field detected, will zero-extend to i64"
-                            )
-                    else:
-                        logger.warning(
-                            f"Unusual field size {field_size_bits} bits, using default 64"
-                        )
-                except Exception as e:
-                    logger.warning(
-                        f"Could not determine field size: {e}, using default 64"
-                    )
-
-            elif field_data.type.__module__ == "vmlinux":
-                # For pointers to structs or complex vmlinux types
-                if field_data.ctype_complex_type is not None and issubclass(
-                    field_data.ctype_complex_type, ctypes._Pointer
-                ):
-                    int_width = 64  # Pointers are always 64-bit
-                    field_size_bytes = 8
-                    logger.info("Field is a pointer type, using 64 bits")
-                else:
-                    logger.warning("Complex vmlinux field type, using default 64 bits")
-
-        # Allocate local storage for the field value
-        local_storage = builder.alloca(ir.IntType(int_width))
-        local_storage_i8_ptr = builder.bitcast(local_storage, i8_ptr_type)
-
-        # Use bpf_probe_read_kernel to safely read the field
-        # This generates:
-        #   %gep = getelementptr i8, ptr %struct_ptr, i64 %offset (already done above as field_ptr)
-        #   %passed = tail call ptr @llvm.bpf.passthrough.p0.p0(i32 2, ptr %gep)
-        #   %result = call i64 inttoptr (i64 113 to ptr)(ptr %local_storage, i32 %size, ptr %passed)
-        from pythonbpf.helper import emit_probe_read_kernel_call
-
-        emit_probe_read_kernel_call(
-            builder, local_storage_i8_ptr, field_size_bytes, field_ptr
-        )
-
-        # Load the value from local storage
-        value = builder.load(local_storage)
-
-        # Zero-extend i32 to i64 if needed
-        if needs_zext:
-            value = builder.zext(value, ir.IntType(64))
-            logger.info("Zero-extended i32 value to i64")
-
-        return value
-
     @staticmethod
     def load_ctx_field(builder, ctx_arg, offset_global, field_data, struct_name=None):
         """

tests/c-form/Makefile

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

tests/c-form/requests.bpf.c

@@ -1,18 +0,0 @@
-#include "vmlinux.h"
-#include <bpf/bpf_helpers.h>
-#include <bpf/bpf_tracing.h>
-#include <bpf/bpf_core_read.h>
-
-char LICENSE[] SEC("license") = "GPL";
-
-SEC("kprobe/blk_mq_start_request")
-int example(struct pt_regs *ctx)
-{
-    u64 a = ctx->r15;
-    struct request *req = (struct request *)(ctx->di);
-    unsigned int something_ns = BPF_CORE_READ(req, timeout);
-    unsigned int data_len = BPF_CORE_READ(req, __data_len);
-    bpf_printk("data length %lld %ld %ld\n", data_len,  something_ns, a);
-
-    return 0;
-}

tests/c-form/requests2.bpf.c

@@ -1,18 +0,0 @@
-#include "vmlinux.h"
-#include <bpf/bpf_helpers.h>
-#include <bpf/bpf_tracing.h>
-#include <bpf/bpf_core_read.h>
-
-char LICENSE[] SEC("license") = "GPL";
-
-SEC("kprobe/blk_mq_start_request")
-int example(struct pt_regs *ctx)
-{
-    u64 a = ctx->r15;
-    struct request *req = (struct request *)(ctx->di);
-    unsigned int something_ns = req->timeout;
-    unsigned int data_len = req->__data_len;
-    bpf_printk("data length %lld %ld %ld\n", data_len,  something_ns, a);
-
-    return 0;
-}

tests/failing_tests/vmlinux/assignment_handling.py

@@ -1,22 +0,0 @@
-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/passing_tests/hash_map_struct.py

@@ -1,42 +0,0 @@
-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/requests.py

@@ -1,27 +0,0 @@
-from vmlinux import struct_request, struct_pt_regs
-from pythonbpf import bpf, section, bpfglobal, compile_to_ir, compile
-import logging
-from ctypes import c_int64
-
-
-@bpf
-@section("kprobe/blk_mq_start_request")
-def example(ctx: struct_pt_regs) -> c_int64:
-    a = ctx.r15
-    req = struct_request(ctx.di)
-    d = req.__data_len
-    b = ctx.r12
-    c = req.timeout
-    print(f"data length {d} and {c} and {a}")
-    print(f"ctx arg {b}")
-    return c_int64(0)
-
-
-@bpf
-@bpfglobal
-def LICENSE() -> str:
-    return "GPL"
-
-
-compile_to_ir("requests.py", "requests.ll", loglevel=logging.INFO)
-compile()

tests/passing_tests/vmlinux/requests2.py

@@ -1,21 +0,0 @@
-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)