Fix typo in process_stmt

pythonbpf/binary_ops.py

@@ -9,6 +9,7 @@ logger: Logger = logging.getLogger(__name__)
 def recursive_dereferencer(var, builder):
     """dereference until primitive type comes out"""
     # TODO: Not worrying about stack overflow for now
+    logger.info(f"Dereferencing {var}, type is {var.type}")
     if isinstance(var.type, ir.PointerType):
         a = builder.load(var)
         return recursive_dereferencer(a, builder)
@@ -18,7 +19,7 @@ def recursive_dereferencer(var, builder):
         raise TypeError(f"Unsupported type for dereferencing: {var.type}")
 
 
-def get_operand_value(operand, module, builder, local_sym_tab):
+def get_operand_value(operand, builder, local_sym_tab):
     """Extract the value from an operand, handling variables and constants."""
     if isinstance(operand, ast.Name):
         if operand.id in local_sym_tab:
@@ -29,14 +30,14 @@ def get_operand_value(operand, module, builder, local_sym_tab):
             return ir.Constant(ir.IntType(64), operand.value)
         raise TypeError(f"Unsupported constant type: {type(operand.value)}")
     elif isinstance(operand, ast.BinOp):
-        return handle_binary_op_impl(operand, module, builder, local_sym_tab)
+        return handle_binary_op_impl(operand, builder, local_sym_tab)
     raise TypeError(f"Unsupported operand type: {type(operand)}")
 
 
-def handle_binary_op_impl(rval, module, builder, local_sym_tab):
+def handle_binary_op_impl(rval, builder, local_sym_tab):
     op = rval.op
-    left = get_operand_value(rval.left, module, builder, local_sym_tab)
-    right = get_operand_value(rval.right, module, builder, local_sym_tab)
+    left = get_operand_value(rval.left, builder, local_sym_tab)
+    right = get_operand_value(rval.right, builder, local_sym_tab)
     logger.info(f"left is {left}, right is {right}, op is {op}")
 
     # Map AST operation nodes to LLVM IR builder methods
@@ -61,6 +62,11 @@ def handle_binary_op_impl(rval, module, builder, local_sym_tab):
         raise SyntaxError("Unsupported binary operation")
 
 
-def handle_binary_op(rval, module, builder, var_name, local_sym_tab):
-    result = handle_binary_op_impl(rval, module, builder, local_sym_tab)
-    builder.store(result, local_sym_tab[var_name].var)
+def handle_binary_op(rval, builder, var_name, local_sym_tab):
+    result = handle_binary_op_impl(rval, builder, local_sym_tab)
+    if var_name and var_name in local_sym_tab:
+        logger.info(
+            f"Storing result {result} into variable {local_sym_tab[var_name].var}"
+        )
+        builder.store(result, local_sym_tab[var_name].var)
+    return result, result.type

pythonbpf/codegen.py

@@ -4,11 +4,7 @@ from .license_pass import license_processing
 from .functions_pass import func_proc
 from .maps import maps_proc
 from .structs import structs_proc
-from .globals_pass import (
-    globals_list_creation,
-    globals_processing,
-    populate_global_symbol_table,
-)
+from .globals_pass import globals_processing
 from .debuginfo import DW_LANG_C11, DwarfBehaviorEnum, DebugInfoGenerator
 import os
 import subprocess
@@ -44,18 +40,15 @@ def processor(source_code, filename, module):
     for func_node in bpf_chunks:
         logger.info(f"Found BPF function/struct: {func_node.name}")
 
-    populate_global_symbol_table(tree, module)
-    license_processing(tree, module)
-    globals_processing(tree, module)
-
     structs_sym_tab = structs_proc(tree, module, bpf_chunks)
     map_sym_tab = maps_proc(tree, module, bpf_chunks)
     func_proc(tree, module, bpf_chunks, map_sym_tab, structs_sym_tab)
 
-    globals_list_creation(tree, module)
+    license_processing(tree, module)
+    globals_processing(tree, module)
 
 
-def compile_to_ir(filename: str, output: str, loglevel=logging.WARNING):
+def compile_to_ir(filename: str, output: str, loglevel=logging.INFO):
     logging.basicConfig(
         level=loglevel, format="%(asctime)s [%(levelname)s] %(name)s: %(message)s"
     )
@@ -128,7 +121,7 @@ def compile_to_ir(filename: str, output: str, loglevel=logging.WARNING):
     return output
 
 
-def compile(loglevel=logging.WARNING) -> bool:
+def compile(loglevel=logging.INFO) -> bool:
     # Look one level up the stack to the caller of this function
     caller_frame = inspect.stack()[1]
     caller_file = Path(caller_frame.filename).resolve()
@@ -161,7 +154,7 @@ def compile(loglevel=logging.WARNING) -> bool:
     return success
 
 
-def BPF(loglevel=logging.WARNING) -> BpfProgram:
+def BPF(loglevel=logging.INFO) -> BpfProgram:
     caller_frame = inspect.stack()[1]
     src = inspect.getsource(caller_frame.frame)
     with tempfile.NamedTemporaryFile(

pythonbpf/functions_pass.py

@@ -233,7 +233,7 @@ def handle_assign(
         else:
             logger.info("Unsupported assignment call function type")
     elif isinstance(rval, ast.BinOp):
-        handle_binary_op(rval, module, builder, var_name, local_sym_tab)
+        handle_binary_op(rval, builder, var_name, local_sym_tab)
     else:
         logger.info("Unsupported assignment value type")
 
@@ -385,24 +385,49 @@ def process_stmt(
         )
     elif isinstance(stmt, ast.Return):
         if stmt.value is None:
-            builder.ret(ir.Constant(ir.IntType(32), 0))
+            builder.ret(ir.Constant(ir.IntType(64), 0))
             did_return = True
         elif (
             isinstance(stmt.value, ast.Call)
             and isinstance(stmt.value.func, ast.Name)
             and len(stmt.value.args) == 1
-            and isinstance(stmt.value.args[0], ast.Constant)
-            and isinstance(stmt.value.args[0].value, int)
         ):
-            call_type = stmt.value.func.id
-            if ctypes_to_ir(call_type) != ret_type:
-                raise ValueError(
-                    "Return type mismatch: expected"
-                    f"{ctypes_to_ir(call_type)}, got {call_type}"
-                )
-            else:
-                builder.ret(ir.Constant(ret_type, stmt.value.args[0].value))
+            if isinstance(stmt.value.args[0], ast.Constant) and isinstance(
+                stmt.value.args[0].value, int
+            ):
+                call_type = stmt.value.func.id
+                if ctypes_to_ir(call_type) != ret_type:
+                    raise ValueError(
+                        "Return type mismatch: expected"
+                        f"{ctypes_to_ir(call_type)}, got {call_type}"
+                    )
+                else:
+                    builder.ret(ir.Constant(ret_type, stmt.value.args[0].value))
+                    did_return = True
+            elif isinstance(stmt.value.args[0], ast.BinOp):
+                # TODO: Should be routed through eval_expr
+                val = handle_binary_op(stmt.value.args[0], builder, None, local_sym_tab)
+                if val is None:
+                    raise ValueError("Failed to evaluate return expression")
+                if val[1] != ret_type:
+                    raise ValueError(
+                        "Return type mismatch: expected " f"{ret_type}, got {val[1]}"
+                    )
+                builder.ret(val[0])
                 did_return = True
+            elif isinstance(stmt.value.args[0], ast.Name):
+                if stmt.value.args[0].id in local_sym_tab:
+                    var = local_sym_tab[stmt.value.args[0].id].var
+                    val = builder.load(var)
+                    if val.type != ret_type:
+                        raise ValueError(
+                            "Return type mismatch: expected"
+                            f"{ret_type}, got {val.type}"
+                        )
+                    builder.ret(val)
+                    did_return = True
+                else:
+                    raise ValueError("Failed to evaluate return expression")
         elif isinstance(stmt.value, ast.Name):
             if stmt.value.id == "XDP_PASS":
                 builder.ret(ir.Constant(ret_type, 2))
@@ -455,6 +480,9 @@ def allocate_mem(
                 continue
             var_name = target.id
             rval = stmt.value
+            if var_name in local_sym_tab:
+                logger.info(f"Variable {var_name} already allocated")
+                continue
             if isinstance(rval, ast.Call):
                 if isinstance(rval.func, ast.Name):
                     call_type = rval.func.id
@@ -568,7 +596,7 @@ def process_func_body(
         )
 
     if not did_return:
-        builder.ret(ir.Constant(ir.IntType(32), 0))
+        builder.ret(ir.Constant(ir.IntType(64), 0))
 
 
 def process_bpf_chunk(func_node, module, return_type, map_sym_tab, structs_sym_tab):

pythonbpf/globals_pass.py

@@ -1,121 +1,8 @@
 from llvmlite import ir
 import ast
 
-from logging import Logger
-import logging
-from .type_deducer import ctypes_to_ir
 
-logger: Logger = logging.getLogger(__name__)
-
-# TODO: this is going to be a huge fuck of a headache in the future.
-global_sym_tab = []
-
-
-def populate_global_symbol_table(tree, module: ir.Module):
-    for node in tree.body:
-        if isinstance(node, ast.FunctionDef):
-            for dec in node.decorator_list:
-                if (
-                    isinstance(dec, ast.Call)
-                    and isinstance(dec.func, ast.Name)
-                    and dec.func.id == "section"
-                    and len(dec.args) == 1
-                    and isinstance(dec.args[0], ast.Constant)
-                    and isinstance(dec.args[0].value, str)
-                ):
-                    global_sym_tab.append(node)
-                elif isinstance(dec, ast.Name) and dec.id == "bpfglobal":
-                    global_sym_tab.append(node)
-
-                elif isinstance(dec, ast.Name) and dec.id == "map":
-                    global_sym_tab.append(node)
-    return False
-
-
-def emit_global(module: ir.Module, node, name):
-    logger.info(f"global identifier {name} processing")
-    # deduce LLVM type from the annotated return
-    if not isinstance(node.returns, ast.Name):
-        raise ValueError(f"Unsupported return annotation {ast.dump(node.returns)}")
-    ty = ctypes_to_ir(node.returns.id)
-
-    # extract the return expression
-    # TODO: turn this return extractor into a generic function I can use everywhere.
-    ret_stmt = node.body[0]
-    if not isinstance(ret_stmt, ast.Return) or ret_stmt.value is None:
-        raise ValueError(f"Global '{name}' has no valid return")
-
-    init_val = ret_stmt.value
-
-    # simple constant like "return 0"
-    if isinstance(init_val, ast.Constant):
-        llvm_init = ir.Constant(ty, init_val.value)
-
-    # variable reference like "return SOME_CONST"
-    elif isinstance(init_val, ast.Name):
-        # need symbol resolution here, stub as 0 for now
-        raise ValueError(f"Name reference {init_val.id} not yet supported")
-
-    # constructor call like "return c_int64(0)" or dataclass(...)
-    elif isinstance(init_val, ast.Call):
-        if len(init_val.args) >= 1 and isinstance(init_val.args[0], ast.Constant):
-            llvm_init = ir.Constant(ty, init_val.args[0].value)
-        else:
-            logger.info("Defaulting to zero as no constant argument found")
-            llvm_init = ir.Constant(ty, 0)
-    else:
-        raise ValueError(f"Unsupported return expr {ast.dump(init_val)}")
-
-    gvar = ir.GlobalVariable(module, ty, name=name)
-    gvar.initializer = llvm_init
-    gvar.align = 8
-    gvar.linkage = "dso_local"
-    gvar.global_constant = False
-    return gvar
-
-
-def globals_processing(tree, module):
-    """Process stuff decorated with @bpf and @bpfglobal except license and return the section name"""
-    globals_sym_tab = []
-
-    for node in tree.body:
-        # Skip non-assignment and non-function nodes
-        if not (isinstance(node, ast.FunctionDef)):
-            continue
-
-        # Get the name based on node type
-        if isinstance(node, ast.FunctionDef):
-            name = node.name
-        else:
-            continue
-
-        # Check for duplicate names
-        if name in globals_sym_tab:
-            raise SyntaxError(f"ERROR: Global name '{name}' previously defined")
-        else:
-            globals_sym_tab.append(name)
-
-        if isinstance(node, ast.FunctionDef) and node.name != "LICENSE":
-            decorators = [
-                dec.id for dec in node.decorator_list if isinstance(dec, ast.Name)
-            ]
-            if "bpf" in decorators and "bpfglobal" in decorators:
-                if (
-                    len(node.body) == 1
-                    and isinstance(node.body[0], ast.Return)
-                    and node.body[0].value is not None
-                    and isinstance(
-                        node.body[0].value, (ast.Constant, ast.Name, ast.Call)
-                    )
-                ):
-                    emit_global(module, node, name)
-                else:
-                    raise SyntaxError(f"ERROR: Invalid syntax for {name} global")
-
-    return None
-
-
-def emit_llvm_compiler_used(module: ir.Module, names: list[str]):
+def emit_globals(module: ir.Module, names: list[str]):
     """
     Emit the @llvm.compiler.used global given a list of function/global names.
     """
@@ -137,7 +24,7 @@ def emit_llvm_compiler_used(module: ir.Module, names: list[str]):
     gv.section = "llvm.metadata"
 
 
-def globals_list_creation(tree, module: ir.Module):
+def globals_processing(tree, module: ir.Module):
     collected = ["LICENSE"]
 
     for node in tree.body:
@@ -153,11 +40,10 @@ def globals_list_creation(tree, module: ir.Module):
                 ):
                     collected.append(node.name)
 
-                # NOTE: all globals other than
-                # elif isinstance(dec, ast.Name) and dec.id == "bpfglobal":
-                #     collected.append(node.name)
+                elif isinstance(dec, ast.Name) and dec.id == "bpfglobal":
+                    collected.append(node.name)
 
                 elif isinstance(dec, ast.Name) and dec.id == "map":
                     collected.append(node.name)
 
-    emit_llvm_compiler_used(module, collected)
+    emit_globals(module, collected)

tests/c-form/globals.bpf.c

@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
-#include <linux/bpf.h>
-#include <bpf/bpf_helpers.h>
-#include <bpf/bpf_tracing.h>
-#include <linux/types.h>
-
-struct test_struct {
-    __u64 a;
-    __u64 b;
-};
-
-struct test_struct w = {};
-volatile __u64 prev_time = 0;
-
-SEC("tracepoint/syscalls/sys_enter_execve")
-int trace_execve(void *ctx)
-{
-    bpf_printk("previous %ul now %ul", w.b, w.a);
-    __u64 ts = bpf_ktime_get_ns();
-    bpf_printk("prev %ul now %ul", prev_time, ts);
-    w.a = ts;
-    w.b = prev_time;
-    prev_time = ts;
-    return 0;
-}
-
-char LICENSE[] SEC("license") = "GPL";

tests/failing_tests/direct_assign.py

@@ -4,6 +4,18 @@ from pythonbpf.maps import HashMap
 
 from ctypes import c_void_p, c_int64
 
+# NOTE: I have decided to not fix this example for now.
+# The issue is in line 31, where we are passing an expression.
+# The update helper expects a pointer type. But the problem is
+# that we must allocate the space for said pointer in the first
+# basic block. As that usage is in a different basic block, we
+# are unable to cast the expression to a pointer type. (as we never
+# allocated space for it).
+# Shall we change our space allocation logic? That allows users to
+# spam the same helper with the same args, and still run out of
+# stack space. So we consider this usage invalid for now.
+# Might fix it later.
+
 
 @bpf
 @map
@@ -14,12 +26,12 @@ def count() -> HashMap:
 @bpf
 @section("xdp")
 def hello_world(ctx: c_void_p) -> c_int64:
-    prev = count().lookup(0)
+    prev = count.lookup(0)
     if prev:
-        count().update(0, prev + 1)
+        count.update(0, prev + 1)
         return XDP_PASS
     else:
-        count().update(0, 1)
+        count.update(0, 1)
 
     return XDP_PASS
 

tests/failing_tests/globals.py

@@ -1,101 +0,0 @@
-import logging
-
-from pythonbpf import compile, bpf, section, bpfglobal, compile_to_ir
-from ctypes import c_void_p, c_int64, c_int32
-
-@bpf
-@bpfglobal
-def somevalue() -> c_int32:
-    return c_int32(42)
-
-@bpf
-@bpfglobal
-def somevalue2() -> c_int64:
-    return c_int64(69)
-
-@bpf
-@bpfglobal
-def somevalue1() -> c_int32:
-    return c_int32(42)
-
-
-# --- Passing examples ---
-
-# Simple constant return
-@bpf
-@bpfglobal
-def g1() -> c_int64:
-    return c_int64(42)
-
-# Constructor with one constant argument
-@bpf
-@bpfglobal
-def g2() -> c_int64:
-    return c_int64(69)
-
-
-# --- Failing examples ---
-
-# No return annotation
-# @bpf
-# @bpfglobal
-# def g3():
-#     return 42
-
-# Return annotation is complex
-# @bpf
-# @bpfglobal
-# def g4() -> List[int]:
-#     return []
-
-# # Return is missing
-# @bpf
-# @bpfglobal
-# def g5() -> c_int64:
-#     pass
-
-# # Return is a variable reference
-# #TODO: maybe fix this sometime later. It defaults to 0
-# CONST = 5
-# @bpf
-# @bpfglobal
-# def g6() -> c_int64:
-#     return c_int64(CONST)
-
-# Constructor with multiple args
-#TODO: this is not working. should it work ?
-@bpf
-@bpfglobal
-def g7() -> c_int64:
-    return c_int64(1)
-
-# Dataclass call
-#TODO: fails with dataclass
-# @dataclass
-# class Point:
-#     x: c_int64
-#     y: c_int64
-
-# @bpf
-# @bpfglobal
-# def g8() -> Point:
-#     return Point(1, 2)
-
-
-@bpf
-@section("tracepoint/syscalls/sys_enter_execve")
-def sometag(ctx: c_void_p) -> c_int64:
-    print("test")
-    global somevalue
-    somevalue = 2
-    print(f"{somevalue}")
-    return c_int64(1)
-
-@bpf
-@bpfglobal
-def LICENSE() -> str:
-    return "GPL"
-
-
-compile_to_ir("globals.py", "globals.ll", loglevel=logging.INFO)
-compile()

tests/failing_tests/named_arg.py

@@ -0,0 +1,40 @@
+from pythonbpf import bpf, map, section, bpfglobal, compile
+from pythonbpf.helper import XDP_PASS
+from pythonbpf.maps import HashMap
+
+from ctypes import c_void_p, c_int64
+
+# NOTE: This example exposes the problems with our typing system.
+# We can't do steps on line 25 and 27.
+# prev is of type i64**. For prev + 1, we deref it down to i64
+# To assign it back to prev, we need to go back to i64**.
+# We cannot allocate space for the intermediate type now.
+# We probably need to track the ref/deref chain for each variable.
+
+@bpf
+@map
+def count() -> HashMap:
+    return HashMap(key=c_int64, value=c_int64, max_entries=1)
+
+
+@bpf
+@section("xdp")
+def hello_world(ctx: c_void_p) -> c_int64:
+    prev = count.lookup(0)
+    if prev:
+        prev = prev + 1
+        count.update(0, prev)
+        return XDP_PASS
+    else:
+        count.update(0, 1)
+
+    return XDP_PASS
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile()

tests/failing_tests/undeclared_values.py

@@ -1,21 +0,0 @@
-import logging
-
-from pythonbpf import compile, bpf, section, bpfglobal, compile_to_ir
-from ctypes import c_void_p, c_int64
-
-# This should not pass as somevalue is not declared at all.
-@bpf
-@section("tracepoint/syscalls/sys_enter_execve")
-def sometag(ctx: c_void_p) -> c_int64:
-    print("test")
-    print(f"{somevalue}")  # noqa: F821
-    return c_int64(1)
-
-@bpf
-@bpfglobal
-def LICENSE() -> str:
-    return "GPL"
-
-
-compile_to_ir("globals.py", "globals.ll", loglevel=logging.INFO)
-compile()

tests/passing_tests/var_rval.py

@@ -0,0 +1,20 @@
+import logging
+
+from pythonbpf import compile, bpf, section, bpfglobal
+from ctypes import c_void_p, c_int64
+
+
+@bpf
+@section("sometag1")
+def sometag(ctx: c_void_p) -> c_int64:
+    a = 1 - 1
+    return c_int64(a)
+
+
+@bpf
+@bpfglobal
+def LICENSE() -> str:
+    return "GPL"
+
+
+compile(loglevel=logging.INFO)