Auto merge of #118284 - RalfJung:miri, r=RalfJung

Miri subtree update

Author: bors

src/tools/miri/.github/workflows/ci.yml

@@ -35,10 +35,6 @@ jobs:
     steps:
       - uses: actions/checkout@v3
 
-      - name: Set the tag GC interval to 1 on linux
-        if: runner.os == 'Linux'
-        run: echo "MIRIFLAGS=-Zmiri-provenance-gc=1" >> $GITHUB_ENV
-
       # Cache the global cargo directory, but NOT the local `target` directory which
       # we cannot reuse anyway when the nightly changes (and it grows quite large
       # over time).

src/tools/miri/cargo-miri/src/setup.rs

@@ -2,6 +2,7 @@
 
 use std::env;
 use std::ffi::OsStr;
+use std::fmt::Write;
 use std::path::PathBuf;
 use std::process::{self, Command};
 
@@ -140,12 +141,20 @@ pub fn setup(
     // Do the build.
     if print_sysroot {
         // Be silent.
-    } else if only_setup {
-        // We want to be explicit.
-        eprintln!("Preparing a sysroot for Miri (target: {target})...");
     } else {
-        // We want to be quiet, but still let the user know that something is happening.
-        eprint!("Preparing a sysroot for Miri (target: {target})... ");
+        let mut msg = String::new();
+        write!(msg, "Preparing a sysroot for Miri (target: {target})").unwrap();
+        if verbose > 0 {
+            write!(msg, " in {}", sysroot_dir.display()).unwrap();
+        }
+        write!(msg, "...").unwrap();
+        if only_setup {
+            // We want to be explicit.
+            eprintln!("{msg}");
+        } else {
+            // We want to be quiet, but still let the user know that something is happening.
+            eprint!("{msg} ");
+        }
     }
     SysrootBuilder::new(&sysroot_dir, target)
         .build_mode(BuildMode::Check)

src/tools/miri/ci.sh

@@ -30,16 +30,23 @@ endgroup
 
 # Test
 function run_tests {
-  if [ -n "${MIRI_TEST_TARGET+exists}" ]; then
+  if [ -n "${MIRI_TEST_TARGET:-}" ]; then
     begingroup "Testing foreign architecture $MIRI_TEST_TARGET"
   else
     begingroup "Testing host architecture"
   fi
 
   ## ui test suite
-  ./miri test
-  if [ -z "${MIRI_TEST_TARGET+exists}" ]; then
-    # Host-only tests: running these on all targets is unlikely to catch more problems and would
+  # On the host and on Linux, also stress-test the GC.
+  if [ -z "${MIRI_TEST_TARGET:-}" ] || [ "$HOST_TARGET" = x86_64-unknown-linux-gnu ]; then
+    MIRIFLAGS="${MIRIFLAGS:-} -Zmiri-provenance-gc=1" ./miri test
+  else
+    ./miri test
+  fi
+
+  # Host-only tests
+  if [ -z "${MIRI_TEST_TARGET:-}" ]; then
+    # Running these on all targets is unlikely to catch more problems and would
     # cost a lot of CI time.
 
     # Tests with optimizations (`-O` is what cargo passes, but crank MIR optimizations up all the
@@ -85,10 +92,11 @@ function run_tests {
 }
 
 function run_tests_minimal {
-  if [ -n "${MIRI_TEST_TARGET+exists}" ]; then
+  if [ -n "${MIRI_TEST_TARGET:-}" ]; then
     begingroup "Testing MINIMAL foreign architecture $MIRI_TEST_TARGET: only testing $@"
   else
-    begingroup "Testing MINIMAL host architecture: only testing $@"
+    echo "run_tests_minimal requires MIRI_TEST_TARGET to be set"
+    exit 1
   fi
 
   ./miri test -- "$@"
@@ -99,16 +107,22 @@ function run_tests_minimal {
   endgroup
 }
 
-# host
+## Main Testing Logic ##
+
+# Host target.
 run_tests
 
+# Extra targets.
+# In particular, fully cover all tier 1 targets.
 case $HOST_TARGET in
   x86_64-unknown-linux-gnu)
     MIRI_TEST_TARGET=i686-unknown-linux-gnu run_tests
     MIRI_TEST_TARGET=aarch64-unknown-linux-gnu run_tests
     MIRI_TEST_TARGET=aarch64-apple-darwin run_tests
     MIRI_TEST_TARGET=i686-pc-windows-gnu run_tests
-    MIRI_TEST_TARGET=x86_64-unknown-freebsd run_tests_minimal hello integer vec panic/panic concurrency/simple pthread-threadname libc-getentropy libc-getrandom libc-misc atomic env align
+    # Some targets are only partially supported.
+    MIRI_TEST_TARGET=x86_64-unknown-freebsd run_tests_minimal hello integer vec panic/panic concurrency/simple pthread-threadname libc-getentropy libc-getrandom libc-misc libc-fs atomic env align
+    MIRI_TEST_TARGET=i686-unknown-freebsd run_tests_minimal hello integer vec panic/panic concurrency/simple pthread-threadname libc-getentropy libc-getrandom libc-misc libc-fs atomic env align
     MIRI_TEST_TARGET=aarch64-linux-android run_tests_minimal hello integer vec panic/panic
     MIRI_TEST_TARGET=wasm32-wasi run_tests_minimal no_std integer strings wasm
     MIRI_TEST_TARGET=wasm32-unknown-unknown run_tests_minimal no_std integer strings wasm

src/tools/miri/miri-script/src/commands.rs

@@ -286,7 +286,7 @@ impl Command {
                 "This will pull a copy of the rust-lang/rust history into this Miri checkout, growing it by about 1GB."
             );
             print!(
-                "To avoid that, abort now and set the `--rustc-git` flag to an existing rustc checkout. Proceed? [y/N] "
+                "To avoid that, abort now and set the `RUSTC_GIT` environment variable to an existing rustc checkout. Proceed? [y/N] "
             );
             std::io::stdout().flush()?;
             let mut answer = String::new();

src/tools/miri/rust-version

@@ -1 +1 @@
-820f06b21f8373060ff7b515715b8440a6a6c197
+3668a8af1b81447c4afa1f82f60d7b94b71a549f

src/tools/miri/src/helpers.rs

@@ -5,17 +5,18 @@ use std::time::Duration;
 
 use log::trace;
 
+use rustc_apfloat::ieee::{Double, Single};
 use rustc_hir::def::{DefKind, Namespace};
 use rustc_hir::def_id::{DefId, CRATE_DEF_INDEX};
 use rustc_index::IndexVec;
 use rustc_middle::mir;
 use rustc_middle::ty::{
     self,
-    layout::{IntegerExt as _, LayoutOf, TyAndLayout},
-    IntTy, Ty, TyCtxt, UintTy,
+    layout::{LayoutOf, TyAndLayout},
+    FloatTy, IntTy, Ty, TyCtxt, UintTy,
 };
 use rustc_span::{def_id::CrateNum, sym, Span, Symbol};
-use rustc_target::abi::{Align, FieldIdx, FieldsShape, Integer, Size, Variants};
+use rustc_target::abi::{Align, FieldIdx, FieldsShape, Size, Variants};
 use rustc_target::spec::abi::Abi;
 
 use rand::RngCore;
@@ -565,10 +566,11 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
     /// is part of the UNIX family. It panics showing a message with the `name` of the foreign function
     /// if this is not the case.
     fn assert_target_os_is_unix(&self, name: &str) {
-        assert!(
-            target_os_is_unix(self.eval_context_ref().tcx.sess.target.os.as_ref()),
-            "`{name}` is only available for supported UNIX family targets",
-        );
+        assert!(self.target_os_is_unix(), "`{name}` is only available for unix targets",);
+    }
+
+    fn target_os_is_unix(&self) -> bool {
+        self.eval_context_ref().tcx.sess.target.families.iter().any(|f| f == "unix")
     }
 
     /// Get last error variable as a place, lazily allocating thread-local storage for it if
@@ -985,65 +987,74 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
         }
     }
 
-    /// Converts `f` to integer type `dest_ty` after rounding with mode `round`.
+    /// Converts `src` from floating point to integer type `dest_ty`
+    /// after rounding with mode `round`.
     /// Returns `None` if `f` is NaN or out of range.
-    fn float_to_int_checked<F>(
+    fn float_to_int_checked(
         &self,
-        f: F,
+        src: &ImmTy<'tcx, Provenance>,
         cast_to: TyAndLayout<'tcx>,
         round: rustc_apfloat::Round,
-    ) -> Option<ImmTy<'tcx, Provenance>>
-    where
-        F: rustc_apfloat::Float + Into<Scalar<Provenance>>,
-    {
+    ) -> InterpResult<'tcx, Option<ImmTy<'tcx, Provenance>>> {
         let this = self.eval_context_ref();
 
-        let val = match cast_to.ty.kind() {
-            // Unsigned
-            ty::Uint(t) => {
-                let size = Integer::from_uint_ty(this, *t).size();
-                let res = f.to_u128_r(size.bits_usize(), round, &mut false);
-                if res.status.intersects(
-                    rustc_apfloat::Status::INVALID_OP
-                        | rustc_apfloat::Status::OVERFLOW
-                        | rustc_apfloat::Status::UNDERFLOW,
-                ) {
-                    // Floating point value is NaN (flagged with INVALID_OP) or outside the range
-                    // of values of the integer type (flagged with OVERFLOW or UNDERFLOW).
-                    return None;
-                } else {
-                    // Floating point value can be represented by the integer type after rounding.
-                    // The INEXACT flag is ignored on purpose to allow rounding.
-                    Scalar::from_uint(res.value, size)
+        fn float_to_int_inner<'tcx, F: rustc_apfloat::Float>(
+            this: &MiriInterpCx<'_, 'tcx>,
+            src: F,
+            cast_to: TyAndLayout<'tcx>,
+            round: rustc_apfloat::Round,
+        ) -> (Scalar<Provenance>, rustc_apfloat::Status) {
+            let int_size = cast_to.layout.size;
+            match cast_to.ty.kind() {
+                // Unsigned
+                ty::Uint(_) => {
+                    let res = src.to_u128_r(int_size.bits_usize(), round, &mut false);
+                    (Scalar::from_uint(res.value, int_size), res.status)
                 }
-            }
-            // Signed
-            ty::Int(t) => {
-                let size = Integer::from_int_ty(this, *t).size();
-                let res = f.to_i128_r(size.bits_usize(), round, &mut false);
-                if res.status.intersects(
-                    rustc_apfloat::Status::INVALID_OP
-                        | rustc_apfloat::Status::OVERFLOW
-                        | rustc_apfloat::Status::UNDERFLOW,
-                ) {
-                    // Floating point value is NaN (flagged with INVALID_OP) or outside the range
-                    // of values of the integer type (flagged with OVERFLOW or UNDERFLOW).
-                    return None;
-                } else {
-                    // Floating point value can be represented by the integer type after rounding.
-                    // The INEXACT flag is ignored on purpose to allow rounding.
-                    Scalar::from_int(res.value, size)
+                // Signed
+                ty::Int(_) => {
+                    let res = src.to_i128_r(int_size.bits_usize(), round, &mut false);
+                    (Scalar::from_int(res.value, int_size), res.status)
                 }
+                // Nothing else
+                _ =>
+                    span_bug!(
+                        this.cur_span(),
+                        "attempted float-to-int conversion with non-int output type {}",
+                        cast_to.ty,
+                    ),
             }
+        }
+
+        let (val, status) = match src.layout.ty.kind() {
+            // f32
+            ty::Float(FloatTy::F32) =>
+                float_to_int_inner::<Single>(this, src.to_scalar().to_f32()?, cast_to, round),
+            // f64
+            ty::Float(FloatTy::F64) =>
+                float_to_int_inner::<Double>(this, src.to_scalar().to_f64()?, cast_to, round),
             // Nothing else
             _ =>
                 span_bug!(
                     this.cur_span(),
-                    "attempted float-to-int conversion with non-int output type {}",
-                    cast_to.ty,
+                    "attempted float-to-int conversion with non-float input type {}",
+                    src.layout.ty,
                 ),
         };
-        Some(ImmTy::from_scalar(val, cast_to))
+
+        if status.intersects(
+            rustc_apfloat::Status::INVALID_OP
+                | rustc_apfloat::Status::OVERFLOW
+                | rustc_apfloat::Status::UNDERFLOW,
+        ) {
+            // Floating point value is NaN (flagged with INVALID_OP) or outside the range
+            // of values of the integer type (flagged with OVERFLOW or UNDERFLOW).
+            Ok(None)
+        } else {
+            // Floating point value can be represented by the integer type after rounding.
+            // The INEXACT flag is ignored on purpose to allow rounding.
+            Ok(Some(ImmTy::from_scalar(val, cast_to)))
+        }
     }
 
     /// Returns an integer type that is twice wide as `ty`
@@ -1063,6 +1074,24 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
             _ => span_bug!(this.cur_span(), "unexpected type: {ty:?}"),
         }
     }
+
+    /// Checks that target feature `target_feature` is enabled.
+    ///
+    /// If not enabled, emits an UB error that states that the feature is
+    /// required by `intrinsic`.
+    fn expect_target_feature_for_intrinsic(
+        &self,
+        intrinsic: Symbol,
+        target_feature: &str,
+    ) -> InterpResult<'tcx, ()> {
+        let this = self.eval_context_ref();
+        if !this.tcx.sess.unstable_target_features.contains(&Symbol::intern(target_feature)) {
+            throw_ub_format!(
+                "attempted to call intrinsic `{intrinsic}` that requires missing target feature {target_feature}"
+            );
+        }
+        Ok(())
+    }
 }
 
 impl<'mir, 'tcx> MiriMachine<'mir, 'tcx> {
@@ -1143,12 +1172,6 @@ pub fn get_local_crates(tcx: TyCtxt<'_>) -> Vec<CrateNum> {
     local_crates
 }
 
-/// Helper function used inside the shims of foreign functions to check that
-/// `target_os` is a supported UNIX OS.
-pub fn target_os_is_unix(target_os: &str) -> bool {
-    matches!(target_os, "linux" | "macos" | "freebsd" | "android")
-}
-
 pub(crate) fn bool_to_simd_element(b: bool, size: Size) -> Scalar<Provenance> {
     // SIMD uses all-1 as pattern for "true". In two's complement,
     // -1 has all its bits set to one and `from_int` will truncate or

src/tools/miri/src/lib.rs

@@ -125,7 +125,7 @@ pub use crate::machine::{
 };
 pub use crate::mono_hash_map::MonoHashMap;
 pub use crate::operator::EvalContextExt as _;
-pub use crate::provenance_gc::{EvalContextExt as _, VisitProvenance, VisitWith, LiveAllocs};
+pub use crate::provenance_gc::{EvalContextExt as _, LiveAllocs, VisitProvenance, VisitWith};
 pub use crate::range_map::RangeMap;
 
 /// Insert rustc arguments at the beginning of the argument list that Miri wants to be

src/tools/miri/src/provenance_gc.rs

@@ -157,15 +157,13 @@ pub struct LiveAllocs<'a, 'mir, 'tcx> {
 
 impl LiveAllocs<'_, '_, '_> {
     pub fn is_live(&self, id: AllocId) -> bool {
-        self.collected.contains(&id) ||
-        self.ecx.is_alloc_live(id)
+        self.collected.contains(&id) || self.ecx.is_alloc_live(id)
     }
 }
 
 impl<'mir, 'tcx: 'mir> EvalContextExt<'mir, 'tcx> for crate::MiriInterpCx<'mir, 'tcx> {}
 pub trait EvalContextExt<'mir, 'tcx: 'mir>: MiriInterpCxExt<'mir, 'tcx> {
     fn run_provenance_gc(&mut self) {
-
         // We collect all tags from various parts of the interpreter, but also
         let this = self.eval_context_mut();
 
@@ -196,10 +194,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: MiriInterpCxExt<'mir, 'tcx> {
 
     fn remove_unreachable_allocs(&mut self, allocs: FxHashSet<AllocId>) {
         let this = self.eval_context_mut();
-        let allocs = LiveAllocs {
-            ecx: this,
-            collected: allocs,
-        };
+        let allocs = LiveAllocs { ecx: this, collected: allocs };
         this.machine.allocation_spans.borrow_mut().retain(|id, _| allocs.is_live(*id));
         this.machine.intptrcast.borrow_mut().remove_unreachable_allocs(&allocs);
         if let Some(borrow_tracker) = &this.machine.borrow_tracker {