angr 9.2.142__py3-none-manylinux2014_aarch64.whl → 9.2.143__py3-none-manylinux2014_aarch64.whl

angr/__init__.py

@@ -2,7 +2,7 @@
 # pylint: disable=wrong-import-position
 from __future__ import annotations
 
-__version__ = "9.2.142"
+__version__ = "9.2.143"
 
 if bytes is str:
     raise Exception(

angr/analyses/calling_convention/calling_convention.py

@@ -220,9 +220,9 @@ class CallingConventionAnalysis(Analysis):
                 self.prototype = prototype  # type: ignore
             return
         if self._function.is_plt:
-            r = self._analyze_plt()
-            if r is not None:
-                self.cc, self.prototype = r
+            r_plt = self._analyze_plt()
+            if r_plt is not None:
+                self.cc, self.prototype, self.prototype_libname = r_plt
             return
 
         r = self._analyze_function()
@@ -278,11 +278,11 @@ class CallingConventionAnalysis(Analysis):
         self.cc = cc
         self.prototype = prototype
 
-    def _analyze_plt(self) -> tuple[SimCC, SimTypeFunction | None] | None:
+    def _analyze_plt(self) -> tuple[SimCC, SimTypeFunction | None, str | None] | None:
         """
         Get the calling convention for a PLT stub.
 
-        :return:    A calling convention.
+        :return:    A calling convention, the function type, as well as the library name if available.
         """
         assert self._function is not None
 
@@ -326,11 +326,11 @@ class CallingConventionAnalysis(Analysis):
                         # we only take the prototype from the SimProcedure if
                         # - the SimProcedure is a function
                         # - the prototype of the SimProcedure is not guessed
-                        return cc, hooker.prototype
+                        return cc, hooker.prototype, hooker.library_name
                 if real_func.prototype is not None:
-                    return cc, real_func.prototype
+                    return cc, real_func.prototype, real_func.prototype_libname
             else:
-                return cc, real_func.prototype
+                return cc, real_func.prototype, real_func.prototype_libname
 
         if self.analyze_callsites:
             # determine the calling convention by analyzing its callsites
@@ -344,7 +344,7 @@ class CallingConventionAnalysis(Analysis):
             prototype = self._adjust_prototype(
                 prototype, callsite_facts, update_arguments=UpdateArgumentsOption.AlwaysUpdate
             )
-            return cc, prototype
+            return cc, prototype, None
 
         return None
 

angr/analyses/calling_convention/fact_collector.py

@@ -1,10 +1,11 @@
 # pylint:disable=too-many-boolean-expressions
 from __future__ import annotations
-from typing import Any
+from typing import Any, TYPE_CHECKING
 
 import pyvex
 import claripy
 
+from angr import SIM_LIBRARIES, SIM_TYPE_COLLECTIONS
 from angr.utils.bits import s2u, u2s
 from angr.block import Block
 from angr.analyses.analysis import Analysis
@@ -13,9 +14,12 @@ from angr.knowledge_plugins.functions import Function
 from angr.codenode import BlockNode, HookNode
 from angr.engines.light import SimEngineNostmtVEX, SimEngineLight, SpOffset, RegisterOffset
 from angr.calling_conventions import SimRegArg, SimStackArg, default_cc
-from angr.sim_type import SimTypeBottom
+from angr.sim_type import SimTypeBottom, dereference_simtype, SimTypeFunction
 from .utils import is_sane_register_variable
 
+if TYPE_CHECKING:
+    from angr.codenode import CodeNode
+
 
 class FactCollectorState:
     """
@@ -224,9 +228,12 @@ class FactCollector(Analysis):
         callee_restored_regs = self._analyze_endpoints_for_restored_regs()
         self._determine_input_args(end_states, callee_restored_regs)
 
-    def _analyze_startpoint(self):
+    def _analyze_startpoint(self) -> list[FactCollectorState]:
         func_graph = self.function.transition_graph
         startpoint = self.function.startpoint
+        if startpoint is None:
+            return []
+
         bp_as_gpr = self.function.info.get("bp_as_gpr", False)
         engine = SimEngineFactCollectorVEX(self.project, bp_as_gpr)
         init_state = FactCollectorState()
@@ -235,9 +242,9 @@ class FactCollector(Analysis):
         init_state.bp_value = init_state.sp_value
 
         traversed = set()
-        queue: list[tuple[int, FactCollectorState, BlockNode | HookNode | Function, BlockNode | HookNode | None]] = [
-            (0, init_state, startpoint, None)
-        ]
+        queue: list[
+            tuple[int, FactCollectorState, CodeNode | BlockNode | HookNode | Function, BlockNode | HookNode | None]
+        ] = [(0, init_state, startpoint, None)]
         end_states: list[FactCollectorState] = []
         while queue:
             depth, state, node, retnode = queue.pop(0)
@@ -398,9 +405,24 @@ class FactCollector(Analysis):
                             and not isinstance(func_succ.prototype.returnty, SimTypeBottom)
                         ):
                             # assume the function overwrites the return variable
-                            returnty_size = func_succ.prototype.returnty.with_arch(self.project.arch).size
-                            assert returnty_size is not None
-                            retval_size = returnty_size // self.project.arch.byte_width
+                            proto = func_succ.prototype
+                            if func_succ.prototype_libname is not None:
+                                # we need to deref the prototype in case it uses SimTypeRef internally
+                                type_collections = []
+                                prototype_lib = SIM_LIBRARIES[func_succ.prototype_libname]
+                                if prototype_lib.type_collection_names:
+                                    for typelib_name in prototype_lib.type_collection_names:
+                                        type_collections.append(SIM_TYPE_COLLECTIONS[typelib_name])
+                                    proto = dereference_simtype(proto, type_collections)
+
+                            assert isinstance(proto, SimTypeFunction) and proto.returnty is not None
+                            returnty_size = proto.returnty.with_arch(self.project.arch).size
+                            if returnty_size is None:
+                                # it may be None if somehow we cannot resolve a SimTypeRef; we fall back to the full
+                                # machine word size
+                                retval_size = self.project.arch.bytes
+                            else:
+                                retval_size = returnty_size // self.project.arch.byte_width
                             retval_sizes.append(retval_size)
                         continue
 

angr/analyses/cfg/indirect_jump_resolvers/const_resolver.py

@@ -43,11 +43,22 @@ class ConstantResolver(IndirectJumpResolver):
     be resolved to a constant value. This resolver must be run after all other more specific resolvers.
     """
 
-    def __init__(self, project):
+    def __init__(self, project, max_func_nodes: int = 512):
         super().__init__(project, timeless=False)
+        self.max_func_nodes = max_func_nodes
 
     def filter(self, cfg, addr, func_addr, block, jumpkind):
+        if not cfg.functions.contains_addr(func_addr):
+            # the function does not exist
+            return False
+
+        # for performance, we don't run constant resolver if the function is too large
+        func = cfg.functions.get_by_addr(func_addr)
+        if len(func.block_addrs_set) > self.max_func_nodes:
+            return False
+
         # we support both an indirect call and jump since the value can be resolved
+
         return jumpkind in {"Ijk_Boring", "Ijk_Call"}
 
     def resolve(  # pylint:disable=unused-argument

angr/analyses/cfg/indirect_jump_resolvers/jumptable.py

@@ -6,6 +6,7 @@ from collections.abc import Sequence
 from collections import defaultdict, OrderedDict
 import logging
 import functools
+import contextlib
 
 import pyvex
 import claripy
@@ -1798,7 +1799,9 @@ class JumpTableResolver(IndirectJumpResolver):
                         # swap the two tmps
                         jump_base_addr.tmp, jump_base_addr.tmp_1 = jump_base_addr.tmp_1, jump_base_addr.tmp
                     # Load the concrete base address
-                    jump_base_addr.base_addr = state.solver.eval(state.scratch.temps[jump_base_addr.tmp_1])
+                    with contextlib.suppress(SimError):
+                        # silently eat the claripy exception
+                        jump_base_addr.base_addr = state.solver.eval(state.scratch.temps[jump_base_addr.tmp_1])
             else:
                 # We do not support the cases where the base address involves more than one addition.
                 # One such case exists in libc-2.27.so shipped with Ubuntu x86 where esi is used as the address of the

angr/analyses/complete_calling_conventions.py

@@ -63,7 +63,7 @@ class CompleteCallingConventionsAnalysis(Analysis):
         max_function_size: int | None = None,
         workers: int = 0,
         cc_callback: Callable | None = None,
-        prioritize_func_addrs: Iterable[int] | None = None,
+        prioritize_func_addrs: list[int] | set[int] | None = None,
         skip_other_funcs: bool = False,
         auto_start: bool = True,
         func_graphs: dict[int, networkx.DiGraph] | None = None,
@@ -130,9 +130,20 @@ class CompleteCallingConventionsAnalysis(Analysis):
         Infer calling conventions for all functions in the current project.
         """
 
-        # get an ordering of functions based on the call graph
-        # note that the call graph is a multi-digraph. we convert it to a digraph to speed up topological sort
-        directed_callgraph = networkx.DiGraph(self.kb.functions.callgraph)
+        # special case: if both _prioritize_func_addrs and _skip_other_funcs are set, we only need to sort part of
+        # the call graph; even better, if there is only one function set, we don't need to sort the call graph at all!
+        if self._prioritize_func_addrs and self._skip_other_funcs:
+            if len(self._prioritize_func_addrs) == 1:
+                self._func_addrs = list(self._prioritize_func_addrs)
+                self._total_funcs = 1
+                return
+            directed_callgraph = networkx.DiGraph(self.kb.functions.callgraph)
+            directed_callgraph = directed_callgraph.subgraph(self._prioritize_func_addrs)
+        else:
+            # get an ordering of functions based on the call graph
+            # note that the call graph is a multi-digraph. we convert it to a digraph to speed up topological sort
+            directed_callgraph = networkx.DiGraph(self.kb.functions.callgraph)
+        assert isinstance(directed_callgraph, networkx.DiGraph)
         sorted_funcs = GraphUtils.quasi_topological_sort_nodes(directed_callgraph)
 
         total_funcs = 0
@@ -148,7 +159,7 @@ class CompleteCallingConventionsAnalysis(Analysis):
                     continue
 
                 if self._max_function_size is not None:
-                    func_size = sum(block.size for block in func.blocks)
+                    func_size = sum(block.size for block in func.blocks if block.size is not None)
                     if func_size > self._max_function_size:
                         _l.info(
                             "Skipping variable recovery for %r since its size (%d) is greater than the cutoff "
@@ -189,6 +200,7 @@ class CompleteCallingConventionsAnalysis(Analysis):
 
     def work(self):
         total_funcs = self._total_funcs
+        assert total_funcs is not None
         if self._workers == 0:
             self._update_progress(0)
             for idx, func_addr in enumerate(self._func_addrs):
@@ -211,6 +223,7 @@ class CompleteCallingConventionsAnalysis(Analysis):
             self._finish_progress()
 
         else:
+            assert self._remaining_funcs is not None and self._func_queue is not None
             self._remaining_funcs.value = len(self._func_addrs)
 
             # generate a call tree (obviously, it's acyclic)

angr/analyses/decompiler/ail_simplifier.py

@@ -36,6 +36,7 @@ from angr.knowledge_plugins.key_definitions.definition import Definition
 from angr.knowledge_plugins.key_definitions.constants import OP_BEFORE
 from angr.errors import AngrRuntimeError
 from angr.analyses import Analysis, AnalysesHub
+from angr.utils.timing import timethis
 from .ailgraph_walker import AILGraphWalker
 from .expression_narrower import ExprNarrowingInfo, NarrowingInfoExtractor, ExpressionNarrower
 from .block_simplifier import BlockSimplifier
@@ -202,6 +203,7 @@ class AILSimplifier(Analysis):
         AILGraphWalker(self.func_graph, _handler, replace_nodes=True).walk()
         self.blocks = {}
 
+    @timethis
     def _compute_reaching_definitions(self) -> SRDAModel:
         # Computing reaching definitions or return the cached one
         if self._reaching_definitions is not None:
@@ -217,6 +219,7 @@ class AILSimplifier(Analysis):
         self._reaching_definitions = rd
         return rd
 
+    @timethis
     def _compute_propagation(self) -> SPropagatorAnalysis:
         # Propagate expressions or return the existing result
         if self._propagator is not None:
@@ -233,6 +236,7 @@ class AILSimplifier(Analysis):
         self._propagator_dead_vvar_ids = prop.dead_vvar_ids
         return prop
 
+    @timethis
     def _compute_equivalence(self) -> set[Equivalence]:
         equivalence = set()
         for block in self.func_graph:
@@ -281,6 +285,7 @@ class AILSimplifier(Analysis):
     # Expression narrowing
     #
 
+    @timethis
     def _narrow_exprs(self) -> bool:
         """
         A register may be used with full width even when only the lower bytes are really needed. This results in the
@@ -511,9 +516,9 @@ class AILSimplifier(Analysis):
             atom = atom_queue.pop(0)
             seen.add(atom)
 
-            use_and_exprs = rd.get_vvar_uses_with_expr(atom)
+            expr_and_uses = rd.all_vvar_uses[atom.varid]
 
-            for loc, expr in use_and_exprs:
+            for expr, loc in set(expr_and_uses):
                 old_block = block_dict.get((loc.block_addr, loc.block_idx), None)
                 if old_block is None:
                     # missing a block for whatever reason
@@ -532,6 +537,7 @@ class AILSimplifier(Analysis):
                     )
                     if new_atom not in seen:
                         atom_queue.append(new_atom)
+                        seen.add(new_atom)
                 else:
                     result.append((atom, loc, expr))
         return result, phi_vars
@@ -659,6 +665,7 @@ class AILSimplifier(Analysis):
     # Unifying local variables
     #
 
+    @timethis
     def _unify_local_variables(self) -> bool:
         """
         Find variables that are definitely equivalent and then eliminate unnecessary copies.
@@ -822,14 +829,14 @@ class AILSimplifier(Analysis):
                             continue
 
                         # find all its uses
-                        all_arg_copy_var_uses: set[tuple[CodeLocation, Any]] = set(
-                            rd.get_vvar_uses_with_expr(arg_copy_def.atom)
+                        all_arg_copy_var_uses: set[tuple[Any, CodeLocation]] = rd.get_vvar_uses_with_expr(
+                            arg_copy_def.atom
                         )
                         all_uses_with_def = set()
 
                         should_abort = False
                         for use in all_arg_copy_var_uses:
-                            used_expr = use[1]
+                            used_expr = use[0]
                             if used_expr is not None and used_expr.size != arg_copy_def.size:
                                 should_abort = True
                                 break
@@ -924,15 +931,19 @@ class AILSimplifier(Analysis):
 
                 # find all uses of this definition
                 # we make a copy of the set since we may touch the set (uses) when replacing expressions
-                all_uses: set[tuple[CodeLocation, Any]] = set(rd.get_vvar_uses_with_expr(to_replace_def.atom))
+                all_uses: set[tuple[Any, CodeLocation]] = set(rd.all_vvar_uses[to_replace_def.atom.varid])
                 # make sure none of these uses are phi nodes (depends on more than one def)
                 all_uses_with_unique_def = set()
-                for use_and_expr in all_uses:
-                    use_loc, used_expr = use_and_expr
+                for expr_and_use in all_uses:
+                    used_expr, use_loc = expr_and_use
                     defs_and_exprs = rd.get_uses_by_location(use_loc, exprs=True)
-                    filtered_defs = {def_ for def_, expr_ in defs_and_exprs if expr_ == used_expr}
+                    filtered_defs = {
+                        def_
+                        for def_, expr_ in defs_and_exprs
+                        if expr_ is not None and used_expr is not None and expr_.varid == used_expr.varid
+                    }
                     if len(filtered_defs) == 1:
-                        all_uses_with_unique_def.add(use_and_expr)
+                        all_uses_with_unique_def.add(expr_and_use)
                     else:
                         # optimization: break early
                         break
@@ -947,7 +958,7 @@ class AILSimplifier(Analysis):
 
                 if not (isinstance(replace_with, VirtualVariable) and replace_with.was_parameter):
                     assignment_ctr = 0
-                    all_use_locs = {use_loc for use_loc, _ in all_uses}
+                    all_use_locs = {use_loc for _, use_loc in all_uses}
                     for use_loc in all_use_locs:
                         if use_loc == eq.codeloc:
                             continue
@@ -960,17 +971,17 @@ class AILSimplifier(Analysis):
                     if assignment_ctr > 1:
                         continue
 
-                all_uses_with_def = {(to_replace_def, use_and_expr) for use_and_expr in all_uses}
+                all_uses_with_def = {(to_replace_def, expr_and_use) for expr_and_use in all_uses}
 
                 remove_initial_assignment = False  # expression folding will take care of it
 
             assert replace_with is not None
 
-            if any(not isinstance(use_and_expr[1], VirtualVariable) for _, use_and_expr in all_uses_with_def):
+            if any(not isinstance(expr_and_use[0], VirtualVariable) for _, expr_and_use in all_uses_with_def):
                 # if any of the uses are phi assignments, we skip
                 used_in_phi_assignment = False
-                for _, use_and_expr in all_uses_with_def:
-                    u = use_and_expr[0]
+                for _, expr_and_use in all_uses_with_def:
+                    u = expr_and_use[1]
                     assert u.block_addr is not None
                     assert u.stmt_idx is not None
                     block = addr_and_idx_to_block[(u.block_addr, u.block_idx)]
@@ -983,8 +994,8 @@ class AILSimplifier(Analysis):
 
             # ensure the uses we consider are all after the eq location
             filtered_all_uses_with_def = []
-            for def_, use_and_expr in all_uses_with_def:
-                u = use_and_expr[0]
+            for def_, expr_and_use in all_uses_with_def:
+                u = expr_and_use[1]
                 if (
                     u.block_addr == eq.codeloc.block_addr
                     and u.block_idx == eq.codeloc.block_idx
@@ -992,7 +1003,7 @@ class AILSimplifier(Analysis):
                 ):
                     # this use happens before the assignment - ignore it
                     continue
-                filtered_all_uses_with_def.append((def_, use_and_expr))
+                filtered_all_uses_with_def.append((def_, expr_and_use))
             all_uses_with_def = filtered_all_uses_with_def
 
             if not all_uses_with_def:
@@ -1004,8 +1015,8 @@ class AILSimplifier(Analysis):
 
             # replace all uses
             all_uses_replaced = True
-            for def_, use_and_expr in all_uses_with_def:
-                u, used_expr = use_and_expr
+            for def_, expr_and_use in all_uses_with_def:
+                used_expr, u = expr_and_use
 
                 use_expr_defns = []
                 for d in rd.get_uses_by_location(u):
@@ -1110,6 +1121,7 @@ class AILSimplifier(Analysis):
         walker.walk_statement(stmt)
         return len(walker.temps) > 0
 
+    @timethis
     def _fold_call_exprs(self) -> bool:
         """
         Fold a call expression (statement) into other statements if the return value of the call expression (statement)
@@ -1183,11 +1195,11 @@ class AILSimplifier(Analysis):
                 assert the_def.codeloc.block_addr is not None
                 assert the_def.codeloc.stmt_idx is not None
 
-                all_uses: set[tuple[CodeLocation, Any]] = set(rd.get_vvar_uses_with_expr(the_def.atom))
+                all_uses: set[tuple[Any, CodeLocation]] = rd.get_vvar_uses_with_expr(the_def.atom)
 
                 if len(all_uses) != 1:
                     continue
-                u, used_expr = next(iter(all_uses))
+                used_expr, u = next(iter(all_uses))
                 if used_expr is None:
                     continue
                 assert u.block_addr is not None
@@ -1314,6 +1326,7 @@ class AILSimplifier(Analysis):
 
         return False, None
 
+    @timethis
     def _iteratively_remove_dead_assignments(self) -> bool:
         anything_removed = False
         while True:
@@ -1323,6 +1336,7 @@ class AILSimplifier(Analysis):
             self._rebuild_func_graph()
             self._clear_cache()
 
+    @timethis
     def _remove_dead_assignments(self) -> bool:
 
         # keeping tracking of statements to remove and statements (as well as dead vvars) to keep allows us to handle
@@ -1330,7 +1344,7 @@ class AILSimplifier(Analysis):
         # value and the floating-point return value.
         stmts_to_remove_per_block: dict[tuple[int, int | None], set[int]] = defaultdict(set)
         stmts_to_keep_per_block: dict[tuple[int, int | None], set[int]] = defaultdict(set)
-        dead_vvar_ids: set[int] = set()
+        dead_vvar_ids: set[int] = self._removed_vvar_ids.copy()
         dead_vvar_codelocs: set[CodeLocation] = set()
         blocks: dict[tuple[int, int | None], Block] = {
             (node.addr, node.idx): self.blocks.get(node, node) for node in self.func_graph.nodes()
@@ -1343,36 +1357,43 @@ class AILSimplifier(Analysis):
         stackarg_offsets = (
             {(tpl[1] & mask) for tpl in self._stack_arg_offsets} if self._stack_arg_offsets is not None else None
         )
+
         while True:
             new_dead_vars_found = False
-            for vvar, codeloc in rd.all_vvar_definitions.items():
-                if vvar.varid in dead_vvar_ids:
+
+            # traverse all virtual variable definitions
+            for vvar_id, codeloc in rd.all_vvar_definitions.items():
+                if vvar_id in dead_vvar_ids:
                     continue
-                if vvar.varid in self._propagator_dead_vvar_ids:
+                uses = None
+                if vvar_id in self._propagator_dead_vvar_ids:
                     # we are definitely removing this variable if it has no uses
-                    uses = rd.all_vvar_uses[vvar]
-                elif vvar.was_stack:
-                    if not self._remove_dead_memdefs:
-                        if rd.is_phi_vvar_id(vvar.varid):
-                            # we always remove unused phi variables
-                            pass
-                        elif vvar.varid in self._secondary_stackvars:
-                            # secondary stack variables are potentially removable
-                            pass
-                        elif stackarg_offsets is not None:
-                            # we always remove definitions for stack arguments
-                            assert vvar.stack_offset is not None
-                            if (vvar.stack_offset & mask) not in stackarg_offsets:
+                    uses = rd.all_vvar_uses[vvar_id]
+
+                if uses is None:
+                    vvar = rd.varid_to_vvar[vvar_id]
+                    if vvar.was_stack:
+                        if not self._remove_dead_memdefs:
+                            if rd.is_phi_vvar_id(vvar_id):
+                                # we always remove unused phi variables
+                                pass
+                            elif vvar_id in self._secondary_stackvars:
+                                # secondary stack variables are potentially removable
+                                pass
+                            elif stackarg_offsets is not None:
+                                # we always remove definitions for stack arguments
+                                assert vvar.stack_offset is not None
+                                if (vvar.stack_offset & mask) not in stackarg_offsets:
+                                    continue
+                            else:
                                 continue
-                        else:
-                            continue
-                    uses = rd.all_vvar_uses[vvar]
+                        uses = rd.all_vvar_uses[vvar_id]
 
-                elif vvar.was_tmp or vvar.was_reg or vvar.was_parameter:
-                    uses = rd.all_vvar_uses[vvar]
+                    elif vvar.was_tmp or vvar.was_reg or vvar.was_parameter:
+                        uses = rd.all_vvar_uses[vvar_id]
 
-                else:
-                    uses = set()
+                    else:
+                        uses = set()
 
                 # remove uses where vvars are going to be removed
                 filtered_uses_count = 0
@@ -1385,7 +1406,7 @@ class AILSimplifier(Analysis):
 
                 if filtered_uses_count == 0:
                     new_dead_vars_found = True
-                    dead_vvar_ids.add(vvar.varid)
+                    dead_vvar_ids.add(vvar_id)
                     dead_vvar_codelocs.add(codeloc)
                     if not isinstance(codeloc, ExternalCodeLocation):
                         assert codeloc.block_addr is not None
@@ -1403,30 +1424,29 @@ class AILSimplifier(Analysis):
                 break
 
         # find all phi variables that rely on variables that no longer exist
-        all_removed_var_ids = self._removed_vvar_ids.copy()
         removed_vvar_ids = self._removed_vvar_ids
         while True:
             new_removed_vvar_ids = set()
             for phi_varid, phi_use_varids in rd.phivarid_to_varids.items():
-                if phi_varid not in all_removed_var_ids and any(
-                    vvarid in removed_vvar_ids for vvarid in phi_use_varids
-                ):
-                    loc = rd.all_vvar_definitions[rd.varid_to_vvar[phi_varid]]
+                if phi_varid not in dead_vvar_ids and any(vvarid in removed_vvar_ids for vvarid in phi_use_varids):
+                    loc = rd.all_vvar_definitions[phi_varid]
                     assert loc.block_addr is not None and loc.stmt_idx is not None
-                    stmts_to_remove_per_block[(loc.block_addr, loc.block_idx)].add(loc.stmt_idx)
-                    new_removed_vvar_ids.add(phi_varid)
-                    all_removed_var_ids.add(phi_varid)
+                    if loc.stmt_idx not in stmts_to_remove_per_block[(loc.block_addr, loc.block_idx)]:
+                        stmts_to_remove_per_block[(loc.block_addr, loc.block_idx)].add(loc.stmt_idx)
+                        new_removed_vvar_ids.add(phi_varid)
+                        dead_vvar_ids.add(phi_varid)
             if not new_removed_vvar_ids:
                 break
             removed_vvar_ids = new_removed_vvar_ids
 
         # find all phi variables that are only ever used by other phi variables
-        redundant_phi_and_dirty_varids = self._find_cyclic_dependent_phis_and_dirty_vvars(rd)
+        redundant_phi_and_dirty_varids = self._find_cyclic_dependent_phis_and_dirty_vvars(rd, dead_vvar_ids)
         for varid in redundant_phi_and_dirty_varids:
-            loc = rd.all_vvar_definitions[rd.varid_to_vvar[varid]]
+            loc = rd.all_vvar_definitions[varid]
             assert loc.block_addr is not None and loc.stmt_idx is not None
-            stmts_to_remove_per_block[(loc.block_addr, loc.block_idx)].add(loc.stmt_idx)
-            stmts_to_keep_per_block[(loc.block_addr, loc.block_idx)].discard(loc.stmt_idx)
+            if loc.stmt_idx not in stmts_to_remove_per_block[(loc.block_addr, loc.block_idx)]:
+                stmts_to_remove_per_block[(loc.block_addr, loc.block_idx)].add(loc.stmt_idx)
+                stmts_to_keep_per_block[(loc.block_addr, loc.block_idx)].discard(loc.stmt_idx)
 
         for codeloc in self._calls_to_remove | self._assignments_to_remove:
             # this call can be removed. make sure it exists in stmts_to_remove_per_block
@@ -1481,6 +1501,7 @@ class AILSimplifier(Analysis):
                         if self._statement_has_call_exprs(stmt):
                             if codeloc in self._calls_to_remove:
                                 # it has a call and must be removed
+                                self._calls_to_remove.discard(codeloc)
                                 simplified = True
                                 continue
                             if isinstance(stmt, Assignment) and isinstance(stmt.dst, VirtualVariable):
@@ -1538,9 +1559,8 @@ class AILSimplifier(Analysis):
         :return:            The set of vvar use atoms.
         """
 
-        vvar = rd.varid_to_vvar[vvar_id]
         used_by: set[int | None] = set()
-        for used_vvar, loc in rd.all_vvar_uses[vvar]:
+        for used_vvar, loc in rd.all_vvar_uses[vvar_id]:
             if used_vvar is None:
                 # no explicit reference
                 used_by.add(None)
@@ -1553,7 +1573,7 @@ class AILSimplifier(Analysis):
                     used_by.add(None)
         return used_by
 
-    def _find_cyclic_dependent_phis_and_dirty_vvars(self, rd: SRDAModel) -> set[int]:
+    def _find_cyclic_dependent_phis_and_dirty_vvars(self, rd: SRDAModel, dead_vvar_ids: set[int]) -> set[int]:
         blocks_dict: dict[tuple[int, int | None], Block] = {(bb.addr, bb.idx): bb for bb in self.func_graph}
 
         # find dirty vvars and vexccall vvars
@@ -1568,16 +1588,21 @@ class AILSimplifier(Analysis):
                 ):
                     dirty_vvar_ids.add(stmt.dst.varid)
 
-        phi_and_dirty_vvar_ids = rd.phi_vvar_ids | dirty_vvar_ids
+        phi_and_dirty_vvar_ids = (rd.phi_vvar_ids | dirty_vvar_ids).difference(dead_vvar_ids)
 
         vvar_used_by: dict[int, set[int | None]] = defaultdict(set)
         for var_id in phi_and_dirty_vvar_ids:
             if var_id in rd.phivarid_to_varids:
                 for used_by_varid in rd.phivarid_to_varids[var_id]:
+                    if used_by_varid in dead_vvar_ids:
+                        # this variable no longer exists
+                        continue
                     if used_by_varid not in vvar_used_by:
-                        vvar_used_by[used_by_varid] |= self._get_vvar_used_by(used_by_varid, rd, blocks_dict)
+                        vvar_used_by[used_by_varid] |= self._get_vvar_used_by(
+                            used_by_varid, rd, blocks_dict
+                        ).difference(dead_vvar_ids)
                     vvar_used_by[used_by_varid].add(var_id)  # probably unnecessary
-            vvar_used_by[var_id] |= self._get_vvar_used_by(var_id, rd, blocks_dict)
+            vvar_used_by[var_id] |= self._get_vvar_used_by(var_id, rd, blocks_dict).difference(dead_vvar_ids)
 
         g = networkx.DiGraph()
         dummy_vvar_id = -1