Add pass to identify undefined or erroneous behaviour

This commit is contained in:
Tomasz Miąsko 2023-12-12 00:00:00 +00:00
parent 920e0051cf
commit 7a246ddd8e
7 changed files with 95 additions and 49 deletions

View File

@ -8,9 +8,6 @@
use rustc_middle::mir::visit::{NonUseContext, PlaceContext, Visitor};
use rustc_middle::mir::*;
use rustc_middle::ty::{self, InstanceDef, ParamEnv, Ty, TyCtxt, TypeVisitableExt, Variance};
use rustc_mir_dataflow::impls::MaybeStorageLive;
use rustc_mir_dataflow::storage::always_storage_live_locals;
use rustc_mir_dataflow::{Analysis, ResultsCursor};
use rustc_target::abi::{Size, FIRST_VARIANT};
use rustc_target::spec::abi::Abi;
@ -51,12 +48,6 @@ fn run_pass(&self, tcx: TyCtxt<'tcx>, body: &mut Body<'tcx>) {
Reveal::All => tcx.param_env_reveal_all_normalized(def_id),
};
let always_live_locals = always_storage_live_locals(body);
let storage_liveness = MaybeStorageLive::new(std::borrow::Cow::Owned(always_live_locals))
.into_engine(tcx, body)
.iterate_to_fixpoint()
.into_results_cursor(body);
let can_unwind = if mir_phase <= MirPhase::Runtime(RuntimePhase::Initial) {
// In this case `AbortUnwindingCalls` haven't yet been executed.
true
@ -83,7 +74,6 @@ fn run_pass(&self, tcx: TyCtxt<'tcx>, body: &mut Body<'tcx>) {
mir_phase,
unwind_edge_count: 0,
reachable_blocks: traversal::reachable_as_bitset(body),
storage_liveness,
place_cache: FxHashSet::default(),
value_cache: FxHashSet::default(),
can_unwind,
@ -116,7 +106,6 @@ struct CfgChecker<'a, 'tcx> {
mir_phase: MirPhase,
unwind_edge_count: usize,
reachable_blocks: BitSet<BasicBlock>,
storage_liveness: ResultsCursor<'a, 'tcx, MaybeStorageLive<'static>>,
place_cache: FxHashSet<PlaceRef<'tcx>>,
value_cache: FxHashSet<u128>,
// If `false`, then the MIR must not contain `UnwindAction::Continue` or
@ -294,28 +283,13 @@ fn is_critical_call_edge(&self, target: Option<BasicBlock>, unwind: UnwindAction
}
impl<'a, 'tcx> Visitor<'tcx> for CfgChecker<'a, 'tcx> {
fn visit_local(&mut self, local: Local, context: PlaceContext, location: Location) {
fn visit_local(&mut self, local: Local, _context: PlaceContext, location: Location) {
if self.body.local_decls.get(local).is_none() {
self.fail(
location,
format!("local {local:?} has no corresponding declaration in `body.local_decls`"),
);
}
if self.reachable_blocks.contains(location.block) && context.is_use() {
// We check that the local is live whenever it is used. Technically, violating this
// restriction is only UB and not actually indicative of not well-formed MIR. This means
// that an optimization which turns MIR that already has UB into MIR that fails this
// check is not necessarily wrong. However, we have no such optimizations at the moment,
// and so we include this check anyway to help us catch bugs. If you happen to write an
// optimization that might cause this to incorrectly fire, feel free to remove this
// check.
self.storage_liveness.seek_after_primary_effect(location);
let locals_with_storage = self.storage_liveness.get();
if !locals_with_storage.contains(local) {
self.fail(location, format!("use of local {local:?}, which has no storage here"));
}
}
}
fn visit_statement(&mut self, statement: &Statement<'tcx>, location: Location) {
@ -367,26 +341,8 @@ fn visit_statement(&mut self, statement: &Statement<'tcx>, location: Location) {
self.fail(location, format!("explicit `{kind:?}` is forbidden"));
}
}
StatementKind::StorageLive(local) => {
// We check that the local is not live when entering a `StorageLive` for it.
// Technically, violating this restriction is only UB and not actually indicative
// of not well-formed MIR. This means that an optimization which turns MIR that
// already has UB into MIR that fails this check is not necessarily wrong. However,
// we have no such optimizations at the moment, and so we include this check anyway
// to help us catch bugs. If you happen to write an optimization that might cause
// this to incorrectly fire, feel free to remove this check.
if self.reachable_blocks.contains(location.block) {
self.storage_liveness.seek_before_primary_effect(location);
let locals_with_storage = self.storage_liveness.get();
if locals_with_storage.contains(*local) {
self.fail(
location,
format!("StorageLive({local:?}) which already has storage here"),
);
}
}
}
StatementKind::StorageDead(_)
StatementKind::StorageLive(_)
| StatementKind::StorageDead(_)
| StatementKind::Intrinsic(_)
| StatementKind::Coverage(_)
| StatementKind::ConstEvalCounter

View File

@ -86,6 +86,7 @@
mod instsimplify;
mod jump_threading;
mod large_enums;
mod lint;
mod lower_intrinsics;
mod lower_slice_len;
mod match_branches;

View File

@ -0,0 +1,77 @@
//! This pass statically detects code which has undefined behaviour or is likely to be erroneous.
//! It can be used to locate problems in MIR building or optimizations. It assumes that all code
//! can be executed, so it has false positives.
use rustc_index::bit_set::BitSet;
use rustc_middle::mir::visit::{PlaceContext, Visitor};
use rustc_middle::mir::*;
use rustc_middle::ty::TyCtxt;
use rustc_mir_dataflow::impls::MaybeStorageLive;
use rustc_mir_dataflow::storage::always_storage_live_locals;
use rustc_mir_dataflow::{Analysis, ResultsCursor};
use std::borrow::Cow;
pub fn lint_body<'tcx>(tcx: TyCtxt<'tcx>, body: &Body<'tcx>, when: String) {
let reachable_blocks = traversal::reachable_as_bitset(body);
let always_live_locals = &always_storage_live_locals(body);
let storage_liveness = MaybeStorageLive::new(Cow::Borrowed(always_live_locals))
.into_engine(tcx, body)
.iterate_to_fixpoint()
.into_results_cursor(body);
Lint { tcx, when, body, reachable_blocks, storage_liveness }.visit_body(body);
}
struct Lint<'a, 'tcx> {
tcx: TyCtxt<'tcx>,
when: String,
body: &'a Body<'tcx>,
reachable_blocks: BitSet<BasicBlock>,
storage_liveness: ResultsCursor<'a, 'tcx, MaybeStorageLive<'a>>,
}
impl<'a, 'tcx> Lint<'a, 'tcx> {
#[track_caller]
fn fail(&self, location: Location, msg: impl AsRef<str>) {
let span = self.body.source_info(location).span;
self.tcx.sess.dcx().span_delayed_bug(
span,
format!(
"broken MIR in {:?} ({}) at {:?}:\n{}",
self.body.source.instance,
self.when,
location,
msg.as_ref()
),
);
}
}
impl<'a, 'tcx> Visitor<'tcx> for Lint<'a, 'tcx> {
fn visit_local(&mut self, local: Local, context: PlaceContext, location: Location) {
if self.reachable_blocks.contains(location.block) && context.is_use() {
self.storage_liveness.seek_after_primary_effect(location);
if !self.storage_liveness.get().contains(local) {
self.fail(location, format!("use of local {local:?}, which has no storage here"));
}
}
}
fn visit_statement(&mut self, statement: &Statement<'tcx>, location: Location) {
match statement.kind {
StatementKind::StorageLive(local) => {
if self.reachable_blocks.contains(location.block) {
self.storage_liveness.seek_before_primary_effect(location);
if self.storage_liveness.get().contains(local) {
self.fail(
location,
format!("StorageLive({local:?}) which already has storage here"),
);
}
}
}
_ => {}
}
self.super_statement(statement, location);
}
}

View File

@ -2,7 +2,7 @@
use rustc_middle::ty::TyCtxt;
use rustc_session::Session;
use crate::{validate, MirPass};
use crate::{lint::lint_body, validate, MirPass};
/// Just like `MirPass`, except it cannot mutate `Body`.
pub trait MirLint<'tcx> {
@ -109,6 +109,7 @@ fn run_passes_inner<'tcx>(
phase_change: Option<MirPhase>,
validate_each: bool,
) {
let lint = tcx.sess.opts.unstable_opts.lint_mir & !body.should_skip();
let validate = validate_each & tcx.sess.opts.unstable_opts.validate_mir & !body.should_skip();
let overridden_passes = &tcx.sess.opts.unstable_opts.mir_enable_passes;
trace!(?overridden_passes);
@ -131,6 +132,9 @@ fn run_passes_inner<'tcx>(
if validate {
validate_body(tcx, body, format!("before pass {name}"));
}
if lint {
lint_body(tcx, body, format!("before pass {name}"));
}
if let Some(prof_arg) = &prof_arg {
tcx.sess
@ -147,6 +151,9 @@ fn run_passes_inner<'tcx>(
if validate {
validate_body(tcx, body, format!("after pass {name}"));
}
if lint {
lint_body(tcx, body, format!("after pass {name}"));
}
body.pass_count += 1;
}
@ -164,6 +171,9 @@ fn run_passes_inner<'tcx>(
if validate || new_phase == MirPhase::Runtime(RuntimePhase::Optimized) {
validate_body(tcx, body, format!("after phase change to {}", new_phase.name()));
}
if lint {
lint_body(tcx, body, format!("after phase change to {}", new_phase.name()));
}
body.pass_count = 1;
}

View File

@ -1694,6 +1694,8 @@ pub(crate) fn parse_function_return(slot: &mut FunctionReturn, v: Option<&str>)
"link native libraries in the linker invocation (default: yes)"),
link_only: bool = (false, parse_bool, [TRACKED],
"link the `.rlink` file generated by `-Z no-link` (default: no)"),
lint_mir: bool = (false, parse_bool, [UNTRACKED],
"lint MIR before and after each transformation"),
llvm_module_flag: Vec<(String, u32, String)> = (Vec::new(), parse_llvm_module_flag, [TRACKED],
"a list of module flags to pass to LLVM (space separated)"),
llvm_plugins: Vec<String> = (Vec::new(), parse_list, [TRACKED],

View File

@ -1,4 +1,4 @@
// compile-flags: -Zvalidate-mir -Ztreat-err-as-bug
// compile-flags: -Zlint-mir -Ztreat-err-as-bug
// failure-status: 101
// error-pattern: broken MIR in
// error-pattern: StorageLive(_1) which already has storage here