miri/

machine.rs

//! Global machine state as well as implementation of the interpreter engine
//! `Machine` trait.

use std::borrow::Cow;
use std::cell::{Cell, RefCell};
use std::collections::BTreeMap;
use std::path::Path;
use std::rc::Rc;
use std::{fmt, process};

use rand::rngs::StdRng;
use rand::{RngExt, SeedableRng};
use rustc_abi::{Align, ExternAbi, Size};
use rustc_apfloat::{Float, FloatConvert};
use rustc_ast::Mutability;
use rustc_ast::expand::allocator::{self, SpecialAllocatorMethod};
use rustc_data_structures::either::Either;
use rustc_data_structures::fx::{FxHashMap, FxHashSet};
#[allow(unused)]
use rustc_data_structures::static_assert_size;
use rustc_hir::attrs::{InlineAttr, Linkage};
use rustc_hir::def::DefKind;
use rustc_log::tracing;
use rustc_middle::middle::codegen_fn_attrs::TargetFeatureKind;
use rustc_middle::mir;
use rustc_middle::query::TyCtxtAt;
use rustc_middle::ty::layout::{
    HasTyCtxt, HasTypingEnv, LayoutCx, LayoutError, LayoutOf, TyAndLayout,
};
use rustc_middle::ty::{self, AtomicOrdering, Instance, Ty, TyCtxt};
use rustc_session::config::InliningThreshold;
use rustc_span::def_id::{CrateNum, DefId};
use rustc_span::{Span, SpanData, Symbol};
use rustc_symbol_mangling::mangle_internal_symbol;
use rustc_target::callconv::FnAbi;
use rustc_target::spec::{Arch, Os};

use crate::alloc_addresses::EvalContextExt;
use crate::concurrency::data_race::{self, NaReadType, NaWriteType};
use crate::concurrency::sync::SyncObj;
use crate::concurrency::{
    AllocDataRaceHandler, GenmcCtx, GenmcEvalContextExt as _, GlobalDataRaceHandler, weak_memory,
};
use crate::helpers::is_no_core;
use crate::*;

/// First real-time signal.
/// `signal(7)` says this must be between 32 and 64 and specifies 34 or 35
/// as typical values.
pub const SIGRTMIN: i32 = 34;

/// Last real-time signal.
/// `signal(7)` says it must be between 32 and 64 and specifies
/// `SIGRTMAX` - `SIGRTMIN` >= 8 (which is the value of `_POSIX_RTSIG_MAX`)
pub const SIGRTMAX: i32 = 42;

/// Each anonymous global (constant, vtable, function pointer, ...) has multiple addresses, but only
/// this many. Since const allocations are never deallocated, choosing a new [`AllocId`] and thus
/// base address for each evaluation would produce unbounded memory usage.
const ADDRS_PER_ANON_GLOBAL: usize = 32;

#[derive(Copy, Clone, Debug, PartialEq)]
pub enum AlignmentCheck {
    /// Do not check alignment.
    None,
    /// Check alignment "symbolically", i.e., using only the requested alignment for an allocation and not its real base address.
    Symbolic,
    /// Check alignment on the actual physical integer address.
    Int,
}

#[derive(Copy, Clone, Debug, PartialEq)]
pub enum RejectOpWith {
    /// Isolated op is rejected with an abort of the machine.
    Abort,

    /// If not Abort, miri returns an error for an isolated op.
    /// Following options determine if user should be warned about such error.
    /// Do not print warning about rejected isolated op.
    NoWarning,

    /// Print a warning about rejected isolated op, with backtrace.
    Warning,

    /// Print a warning about rejected isolated op, without backtrace.
    WarningWithoutBacktrace,
}

#[derive(Copy, Clone, Debug, PartialEq)]
pub enum IsolatedOp {
    /// Reject an op requiring communication with the host. By
    /// default, miri rejects the op with an abort. If not, it returns
    /// an error code, and prints a warning about it. Warning levels
    /// are controlled by `RejectOpWith` enum.
    Reject(RejectOpWith),

    /// Execute op requiring communication with the host, i.e. disable isolation.
    Allow,
}

#[derive(Debug, Copy, Clone, PartialEq, Eq)]
pub enum BacktraceStyle {
    /// Prints a terser backtrace which ideally only contains relevant information.
    Short,
    /// Prints a backtrace with all possible information.
    Full,
    /// Prints only the frame that the error occurs in.
    Off,
}

#[derive(Debug, Copy, Clone, PartialEq, Eq)]
pub enum ValidationMode {
    /// Do not perform any kind of validation.
    No,
    /// Validate the interior of the value, but not things behind references.
    Shallow,
    /// Fully recursively validate references.
    Deep,
}

#[derive(Debug, Copy, Clone, PartialEq, Eq)]
pub enum FloatRoundingErrorMode {
    /// Apply a random error (the default).
    Random,
    /// Don't apply any error.
    None,
    /// Always apply the maximum error (with a random sign).
    Max,
}

/// Extra data stored with each stack frame
pub struct FrameExtra<'tcx> {
    /// Extra data for the Borrow Tracker.
    pub borrow_tracker: Option<borrow_tracker::FrameState>,

    /// If this is Some(), then this is a special "catch unwind" frame (the frame of `try_fn`
    /// called by `try`). When this frame is popped during unwinding a panic,
    /// we stop unwinding, use the `CatchUnwindData` to handle catching.
    pub catch_unwind: Option<CatchUnwindData<'tcx>>,

    /// If `measureme` profiling is enabled, holds timing information
    /// for the start of this frame. When we finish executing this frame,
    /// we use this to register a completed event with `measureme`.
    pub timing: Option<measureme::DetachedTiming>,

    /// Indicates how user-relevant this frame is. `#[track_caller]` frames are never relevant.
    /// Frames from user-relevant crates are maximally relevant; frames from other crates are less
    /// relevant.
    pub user_relevance: u8,

    /// Data race detector per-frame data.
    pub data_race: Option<data_race::FrameState>,
}

impl<'tcx> std::fmt::Debug for FrameExtra<'tcx> {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        // Omitting `timing`, it does not support `Debug`.
        let FrameExtra { borrow_tracker, catch_unwind, timing: _, user_relevance, data_race } =
            self;
        f.debug_struct("FrameData")
            .field("borrow_tracker", borrow_tracker)
            .field("catch_unwind", catch_unwind)
            .field("user_relevance", user_relevance)
            .field("data_race", data_race)
            .finish()
    }
}

impl VisitProvenance for FrameExtra<'_> {
    fn visit_provenance(&self, visit: &mut VisitWith<'_>) {
        let FrameExtra { catch_unwind, borrow_tracker, timing: _, user_relevance: _, data_race: _ } =
            self;

        catch_unwind.visit_provenance(visit);
        borrow_tracker.visit_provenance(visit);
    }
}

/// Extra memory kinds
#[derive(Debug, Copy, Clone, PartialEq, Eq)]
pub enum MiriMemoryKind {
    /// `__rust_alloc` memory.
    Rust,
    /// `miri_alloc` memory.
    Miri,
    /// `malloc` memory.
    C,
    /// Windows `HeapAlloc` memory.
    WinHeap,
    /// Windows "local" memory (to be freed with `LocalFree`)
    WinLocal,
    /// Memory for args, errno, env vars, and other parts of the machine-managed environment.
    /// This memory may leak.
    Machine,
    /// Memory allocated by the runtime, e.g. for readdir. Separate from `Machine` because we clean
    /// it up (or expect the user to invoke operations that clean it up) and leak-check it.
    Runtime,
    /// Globals copied from `tcx`.
    /// This memory may leak.
    Global,
    /// Memory for extern statics.
    /// This memory may leak.
    ExternStatic,
    /// Memory for thread-local statics.
    /// This memory may leak.
    Tls,
    /// Memory mapped directly by the program.
    Mmap,
    /// Memory allocated for `getaddrinfo` result.
    SocketAddress,
}

impl From<MiriMemoryKind> for MemoryKind {
    #[inline(always)]
    fn from(kind: MiriMemoryKind) -> MemoryKind {
        MemoryKind::Machine(kind)
    }
}

impl MayLeak for MiriMemoryKind {
    #[inline(always)]
    fn may_leak(self) -> bool {
        use self::MiriMemoryKind::*;
        match self {
            Rust | Miri | C | WinHeap | WinLocal | Runtime => false,
            Machine | Global | ExternStatic | Tls | Mmap | SocketAddress => true,
        }
    }
}

impl MiriMemoryKind {
    /// Whether we have a useful allocation span for an allocation of this kind.
    fn should_save_allocation_span(self) -> bool {
        use self::MiriMemoryKind::*;
        match self {
            // Heap allocations are fine since the `Allocation` is created immediately.
            Rust | Miri | C | WinHeap | WinLocal | Mmap => true,
            // Everything else is unclear, let's not show potentially confusing spans.
            Machine | Global | ExternStatic | Tls | Runtime | SocketAddress => false,
        }
    }
}

impl fmt::Display for MiriMemoryKind {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        use self::MiriMemoryKind::*;
        match self {
            Rust => write!(f, "Rust heap"),
            Miri => write!(f, "Miri bare-metal heap"),
            C => write!(f, "C heap"),
            WinHeap => write!(f, "Windows heap"),
            WinLocal => write!(f, "Windows local memory"),
            Machine => write!(f, "machine-managed memory"),
            Runtime => write!(f, "language runtime memory"),
            Global => write!(f, "global (static or const)"),
            ExternStatic => write!(f, "extern static"),
            Tls => write!(f, "thread-local static"),
            Mmap => write!(f, "mmap"),
            SocketAddress => write!(f, "socket address"),
        }
    }
}

pub type MemoryKind = interpret::MemoryKind<MiriMemoryKind>;

/// Pointer provenance.
// This needs to be `Eq`+`Hash` because the `Machine` trait needs that because validity checking
// *might* be recursive and then it has to track which places have already been visited.
// These implementations are a bit questionable, and it means we may check the same place multiple
// times with different provenance, but that is in general not wrong.
#[derive(Clone, Copy, PartialEq, Eq, Hash)]
pub enum Provenance {
    /// For pointers with concrete provenance. we exactly know which allocation they are attached to
    /// and what their borrow tag is.
    Concrete {
        alloc_id: AllocId,
        /// Borrow Tracker tag.
        tag: BorTag,
    },
    /// Pointers with wildcard provenance are created on int-to-ptr casts. According to the
    /// specification, we should at that point angelically "guess" a provenance that will make all
    /// future uses of this pointer work, if at all possible. Of course such a semantics cannot be
    /// actually implemented in Miri. So instead, we approximate this, erroring on the side of
    /// accepting too much code rather than rejecting correct code: a pointer with wildcard
    /// provenance "acts like" any previously exposed pointer. Each time it is used, we check
    /// whether *some* exposed pointer could have done what we want to do, and if the answer is yes
    /// then we allow the access. This allows too much code in two ways:
    /// - The same wildcard pointer can "take the role" of multiple different exposed pointers on
    ///   subsequent memory accesses.
    /// - In the aliasing model, we don't just have to know the borrow tag of the pointer used for
    ///   the access, we also have to update the aliasing state -- and that update can be very
    ///   different depending on which borrow tag we pick! Stacked Borrows has support for this by
    ///   switching to a stack that is only approximately known, i.e. we over-approximate the effect
    ///   of using *any* exposed pointer for this access, and only keep information about the borrow
    ///   stack that would be true with all possible choices.
    Wildcard,
}

/// The "extra" information a pointer has over a regular AllocId.
#[derive(Copy, Clone, PartialEq)]
pub enum ProvenanceExtra {
    Concrete(BorTag),
    Wildcard,
}

#[cfg(target_pointer_width = "64")]
static_assert_size!(StrictPointer, 24);
// Pointer does not fit as the layout algorithm isn't smart enough (but also, we tried using
// pattern types to get a larger niche that makes this fit and it didn't improve performance).
// #[cfg(target_pointer_width = "64")]
//static_assert_size!(Pointer, 24);
#[cfg(target_pointer_width = "64")]
static_assert_size!(Scalar, 32);

impl fmt::Debug for Provenance {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            Provenance::Concrete { alloc_id, tag } => {
                // Forward `alternate` flag to `alloc_id` printing.
                if f.alternate() {
                    write!(f, "[{alloc_id:#?}]")?;
                } else {
                    write!(f, "[{alloc_id:?}]")?;
                }
                // Print Borrow Tracker tag.
                write!(f, "{tag:?}")?;
            }
            Provenance::Wildcard => {
                write!(f, "[wildcard]")?;
            }
        }
        Ok(())
    }
}

impl interpret::Provenance for Provenance {
    /// We use absolute addresses in the `offset` of a `StrictPointer`.
    const OFFSET_IS_ADDR: bool = true;

    /// Miri implements wildcard provenance.
    const WILDCARD: Option<Self> = Some(Provenance::Wildcard);

    fn get_alloc_id(self) -> Option<AllocId> {
        match self {
            Provenance::Concrete { alloc_id, .. } => Some(alloc_id),
            Provenance::Wildcard => None,
        }
    }

    fn fmt(ptr: &interpret::Pointer<Self>, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        let (prov, addr) = ptr.into_raw_parts(); // offset is absolute address
        write!(f, "{:#x}", addr.bytes())?;
        if f.alternate() {
            write!(f, "{prov:#?}")?;
        } else {
            write!(f, "{prov:?}")?;
        }
        Ok(())
    }
}

impl fmt::Debug for ProvenanceExtra {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            ProvenanceExtra::Concrete(pid) => write!(f, "{pid:?}"),
            ProvenanceExtra::Wildcard => write!(f, "<wildcard>"),
        }
    }
}

impl ProvenanceExtra {
    pub fn and_then<T>(self, f: impl FnOnce(BorTag) -> Option<T>) -> Option<T> {
        match self {
            ProvenanceExtra::Concrete(pid) => f(pid),
            ProvenanceExtra::Wildcard => None,
        }
    }
}

/// Extra per-allocation data
#[derive(Debug)]
pub struct AllocExtra<'tcx> {
    /// Global state of the borrow tracker, if enabled.
    pub borrow_tracker: Option<borrow_tracker::AllocState>,
    /// Extra state for data race detection.
    ///
    /// Invariant: The enum variant must match the enum variant in the `data_race` field on `MiriMachine`
    pub data_race: AllocDataRaceHandler,
    /// A backtrace to where this allocation was allocated.
    /// As this is recorded for leak reports, it only exists
    /// if this allocation is leakable. The backtrace is not
    /// pruned yet; that should be done before printing it.
    pub backtrace: Option<Vec<FrameInfo<'tcx>>>,
    /// Synchronization objects like to attach extra data to particular addresses. We store that
    /// inside the relevant allocation, to ensure that everything is removed when the allocation is
    /// freed.
    /// This maps offsets to synchronization-primitive-specific data.
    pub sync_objs: BTreeMap<Size, Box<dyn SyncObj>>,
}

// We need a `Clone` impl because the machine passes `Allocation` through `Cow`...
// but that should never end up actually cloning our `AllocExtra`.
impl<'tcx> Clone for AllocExtra<'tcx> {
    fn clone(&self) -> Self {
        panic!("our allocations should never be cloned");
    }
}

impl VisitProvenance for AllocExtra<'_> {
    fn visit_provenance(&self, visit: &mut VisitWith<'_>) {
        let AllocExtra { borrow_tracker, data_race, backtrace: _, sync_objs: _ } = self;

        borrow_tracker.visit_provenance(visit);
        data_race.visit_provenance(visit);
    }
}

/// Precomputed layouts of primitive types
pub struct PrimitiveLayouts<'tcx> {
    pub unit: TyAndLayout<'tcx>,
    pub i8: TyAndLayout<'tcx>,
    pub i16: TyAndLayout<'tcx>,
    pub i32: TyAndLayout<'tcx>,
    pub i64: TyAndLayout<'tcx>,
    pub i128: TyAndLayout<'tcx>,
    pub isize: TyAndLayout<'tcx>,
    pub u8: TyAndLayout<'tcx>,
    pub u16: TyAndLayout<'tcx>,
    pub u32: TyAndLayout<'tcx>,
    pub u64: TyAndLayout<'tcx>,
    pub u128: TyAndLayout<'tcx>,
    pub usize: TyAndLayout<'tcx>,
    pub bool: TyAndLayout<'tcx>,
    pub unit_ptr_mut: TyAndLayout<'tcx>,   // *mut ()
    pub unit_ptr_const: TyAndLayout<'tcx>, // *const ()
    pub void_ptr_mut: TyAndLayout<'tcx>,   // *mut c_void
    pub void_ptr_const: TyAndLayout<'tcx>, // *const c_void
    pub fn_ptr: TyAndLayout<'tcx>,         // extern "C" fn()
}

impl<'tcx> PrimitiveLayouts<'tcx> {
    fn new(layout_cx: LayoutCx<'tcx>) -> Result<Self, &'tcx LayoutError<'tcx>> {
        let tcx = layout_cx.tcx();

        let unit_ptr_mut = Ty::new_mut_ptr(tcx, tcx.types.unit);
        let unit_ptr_const = Ty::new_imm_ptr(tcx, tcx.types.unit);
        // We fall back to `()` if the lang item is missing, so `no_core` works better with Miri.
        let c_void = match tcx.lang_items().c_void() {
            Some(c_void) => ty::Instance::mono(tcx, c_void).ty(tcx, layout_cx.typing_env),
            None => tcx.types.unit,
        };
        let void_ptr_mut = Ty::new_mut_ptr(tcx, c_void);
        let void_ptr_const = Ty::new_imm_ptr(tcx, c_void);

        let sig_kind = ty::FnSigKind::default()
            .set_abi(ExternAbi::C { unwind: false })
            .set_safety(rustc_hir::Safety::Safe);
        let fn_ptr =
            Ty::new_fn_ptr(tcx, ty::Binder::dummy(tcx.mk_fn_sig([], tcx.types.unit, sig_kind)));

        Ok(Self {
            unit: layout_cx.layout_of(tcx.types.unit)?,
            i8: layout_cx.layout_of(tcx.types.i8)?,
            i16: layout_cx.layout_of(tcx.types.i16)?,
            i32: layout_cx.layout_of(tcx.types.i32)?,
            i64: layout_cx.layout_of(tcx.types.i64)?,
            i128: layout_cx.layout_of(tcx.types.i128)?,
            isize: layout_cx.layout_of(tcx.types.isize)?,
            u8: layout_cx.layout_of(tcx.types.u8)?,
            u16: layout_cx.layout_of(tcx.types.u16)?,
            u32: layout_cx.layout_of(tcx.types.u32)?,
            u64: layout_cx.layout_of(tcx.types.u64)?,
            u128: layout_cx.layout_of(tcx.types.u128)?,
            usize: layout_cx.layout_of(tcx.types.usize)?,
            bool: layout_cx.layout_of(tcx.types.bool)?,
            unit_ptr_mut: layout_cx.layout_of(unit_ptr_mut)?,
            unit_ptr_const: layout_cx.layout_of(unit_ptr_const)?,
            void_ptr_mut: layout_cx.layout_of(void_ptr_mut)?,
            void_ptr_const: layout_cx.layout_of(void_ptr_const)?,
            fn_ptr: layout_cx.layout_of(fn_ptr)?,
        })
    }

    pub fn uint(&self, size: Size) -> Option<TyAndLayout<'tcx>> {
        match size.bits() {
            8 => Some(self.u8),
            16 => Some(self.u16),
            32 => Some(self.u32),
            64 => Some(self.u64),
            128 => Some(self.u128),
            _ => None,
        }
    }

    pub fn int(&self, size: Size) -> Option<TyAndLayout<'tcx>> {
        match size.bits() {
            8 => Some(self.i8),
            16 => Some(self.i16),
            32 => Some(self.i32),
            64 => Some(self.i64),
            128 => Some(self.i128),
            _ => None,
        }
    }
}

/// The machine itself.
///
/// If you add anything here that stores machine values, remember to update
/// `visit_all_machine_values`!
pub struct MiriMachine<'tcx> {
    // We carry a copy of the global `TyCtxt` for convenience, so methods taking just `&Evaluator` have `tcx` access.
    pub tcx: TyCtxt<'tcx>,

    /// Global data for borrow tracking.
    pub borrow_tracker: Option<borrow_tracker::GlobalState>,

    /// Depending on settings, this will be `None`,
    /// global data for a data race detector,
    /// or the context required for running in GenMC mode.
    ///
    /// Invariant: The enum variant must match the enum variant of `AllocDataRaceHandler` in the `data_race` field of all `AllocExtra`.
    pub data_race: GlobalDataRaceHandler,

    /// Ptr-int-cast module global data.
    pub alloc_addresses: alloc_addresses::GlobalState,

    /// Environment variables.
    pub(crate) env_vars: EnvVars<'tcx>,

    /// Return place of the main function.
    pub(crate) main_fn_ret_place: Option<MPlaceTy<'tcx>>,

    /// Program arguments (`Option` because we can only initialize them after creating the ecx).
    /// These are *pointers* to argc/argv because macOS.
    /// We also need the full command line as one string because of Windows.
    pub(crate) argc: Option<Pointer>,
    pub(crate) argv: Option<Pointer>,
    pub(crate) cmd_line: Option<Pointer>,

    /// TLS state.
    pub(crate) tls: TlsData<'tcx>,

    /// What should Miri do when an op requires communicating with the host,
    /// such as accessing host env vars, random number generation, and
    /// file system access.
    pub(crate) isolated_op: IsolatedOp,

    /// Whether to enforce the validity invariant.
    pub(crate) validation: ValidationMode,

    /// The table of file descriptors.
    pub(crate) fds: shims::FdTable,
    /// The table of directory descriptors.
    pub(crate) dirs: shims::DirTable,

    /// Managing file descriptors whose readiness needs to be updated.
    pub(crate) delayed_readiness_updates: Rc<shims::DelayedReadinessUpdates>,

    /// This machine's monotone clock.
    pub(crate) monotonic_clock: MonotonicClock,

    /// The set of threads.
    pub(crate) threads: ThreadManager<'tcx>,

    /// Handles blocking I/O and polling for completion.
    pub(crate) blocking_io: BlockingIoManager,

    /// Stores which thread is eligible to run on which CPUs.
    /// This has no effect at all, it is just tracked to produce the correct result
    /// in `sched_getaffinity`
    /// This will be `None` when running `#![no_core]` crates.
    pub(crate) thread_cpu_affinity: Option<FxHashMap<ThreadId, shims::CpuAffinityMask>>,

    /// Precomputed `TyLayout`s for primitive data types that are commonly used inside Miri.
    pub(crate) layouts: PrimitiveLayouts<'tcx>,

    /// Allocations that are considered roots of static memory (that may leak).
    pub(crate) static_roots: Vec<AllocId>,

    /// The `measureme` profiler used to record timing information about
    /// the emulated program.
    profiler: Option<measureme::Profiler>,
    /// Used with `profiler` to cache the `StringId`s for event names
    /// used with `measureme`.
    string_cache: FxHashMap<String, measureme::StringId>,

    /// Cache of `Instance` exported under the given `Symbol` name.
    /// `None` means no `Instance` exported under the given name is found.
    pub(crate) exported_symbols_cache: RefCell<FxHashMap<Symbol, Option<Instance<'tcx>>>>,

    /// Equivalent setting as RUST_BACKTRACE on encountering an error.
    pub(crate) backtrace_style: BacktraceStyle,

    /// Crates which are considered user-relevant for the purposes of error reporting.
    pub(crate) user_relevant_crates: Vec<CrateNum>,

    /// Mapping extern static names to their pointer.
    pub(crate) extern_statics: FxHashMap<Symbol, StrictPointer>,
    /// Statics with `import_linkage` have an extra indirection
    /// (<https://github.com/rust-lang/rust/issues/156468>) so we keep them in a separate table.
    pub(crate) extern_statics_imports: FxHashMap<Symbol, StrictPointer>,
    /// A pointer to the allocation we provide for non-existent weak symbols.
    pub(crate) extern_static_weak_import_default: Option<StrictPointer>,

    /// The random number generator used for resolving non-determinism.
    /// Needs to be queried by ptr_to_int, hence needs interior mutability.
    pub(crate) rng: RefCell<StdRng>,

    /// The allocator used for the machine's `AllocBytes` in native-libs mode.
    pub(crate) allocator: Option<Rc<RefCell<crate::alloc::isolated_alloc::IsolatedAlloc>>>,

    /// The allocation IDs to report when they are being allocated
    /// (helps for debugging memory leaks and use after free bugs).
    pub(crate) tracked_alloc_ids: FxHashSet<AllocId>,
    /// For the tracked alloc ids, also report read/write accesses.
    track_alloc_accesses: bool,

    /// Controls whether alignment of memory accesses is being checked.
    pub(crate) check_alignment: AlignmentCheck,

    /// Failure rate of compare_exchange_weak, between 0.0 and 1.0
    pub(crate) cmpxchg_weak_failure_rate: f64,

    /// The probability of the active thread being preempted at the end of each basic block.
    pub(crate) preemption_rate: f64,

    /// If `Some`, we will report the current stack every N basic blocks.
    pub(crate) report_progress: Option<u32>,
    // The total number of blocks that have been executed.
    pub(crate) basic_block_count: u64,

    /// Handle of the optional shared object file for native functions.
    #[cfg(all(feature = "native-lib", unix))]
    pub native_lib: Vec<(libloading::Library, std::path::PathBuf)>,
    #[cfg(not(all(feature = "native-lib", unix)))]
    pub native_lib: Vec<!>,
    /// A memory location for exchanging the current `ecx` pointer with native code.
    #[cfg(all(feature = "native-lib", unix))]
    pub native_lib_ecx_interchange: &'static Cell<usize>,

    /// Run a garbage collector for BorTags every N basic blocks.
    pub(crate) gc_interval: u32,
    /// The number of blocks that passed since the last BorTag GC pass.
    pub(crate) since_gc: u32,

    /// The number of CPUs to be reported by miri.
    pub(crate) num_cpus: u32,

    /// Determines Miri's page size and associated values
    pub(crate) page_size: u64,
    pub(crate) stack_addr: u64,
    pub(crate) stack_size: u64,

    /// Whether to collect a backtrace when each allocation is created, just in case it leaks.
    pub(crate) collect_leak_backtraces: bool,

    /// The spans we will use to report where an allocation was created and deallocated in
    /// diagnostics.
    pub(crate) allocation_spans: RefCell<FxHashMap<AllocId, (Span, Option<Span>)>>,

    /// For each allocation, an offset inside that allocation that was deemed aligned even for
    /// symbolic alignment checks. This cannot be stored in `AllocExtra` since it needs to be
    /// tracked for vtables and function allocations as well as regular allocations.
    ///
    /// Invariant: the promised alignment will never be less than the native alignment of the
    /// allocation.
    pub(crate) symbolic_alignment: RefCell<FxHashMap<AllocId, (Size, Align)>>,

    /// A cache of "data range" computations for unions (i.e., the offsets of non-padding bytes).
    union_data_ranges: FxHashMap<Ty<'tcx>, RangeSet>,

    /// Caches the sanity-checks for various pthread primitives.
    pub(crate) pthread_mutex_sanity: Cell<bool>,
    pub(crate) pthread_rwlock_sanity: Cell<bool>,
    pub(crate) pthread_condvar_sanity: Cell<bool>,

    /// (Foreign) symbols that are synthesized as part of the allocator shim: the key indicates the
    /// name of the symbol being synthesized; the value indicates whether this should invoke some
    /// other symbol or whether this has special allocator semantics.
    pub(crate) allocator_shim_symbols: FxHashMap<Symbol, Either<Symbol, SpecialAllocatorMethod>>,
    /// Cache for `mangle_internal_symbol`.
    pub(crate) mangle_internal_symbol_cache: FxHashMap<&'static str, String>,

    /// Whether floating-point operations can behave non-deterministically.
    pub float_nondet: bool,
    /// Whether floating-point operations can have a non-deterministic rounding error.
    pub float_rounding_error: FloatRoundingErrorMode,

    /// Whether Miri artificially introduces short reads/writes on file descriptors.
    pub short_fd_operations: bool,
}

impl<'tcx> MiriMachine<'tcx> {
    /// Create a new MiriMachine.
    ///
    /// Invariant: `genmc_ctx.is_some() == config.genmc_config.is_some()`
    pub(crate) fn new(
        config: &MiriConfig,
        layout_cx: LayoutCx<'tcx>,
        genmc_ctx: Option<Rc<GenmcCtx>>,
    ) -> Self {
        let tcx = layout_cx.tcx();
        let user_relevant_crates = Self::get_user_relevant_crates(tcx, config);
        let layouts =
            PrimitiveLayouts::new(layout_cx).expect("Couldn't get layouts of primitive types");
        let profiler = config.measureme_out.as_ref().map(|out| {
            let crate_name =
                tcx.sess.opts.crate_name.clone().unwrap_or_else(|| "unknown-crate".to_string());
            let pid = process::id();
            // We adopt the same naming scheme for the profiler output that rustc uses. In rustc,
            // the PID is padded so that the nondeterministic value of the PID does not spread
            // nondeterminism to the allocator. In Miri we are not aiming for such performance
            // control, we just pad for consistency with rustc.
            let filename = format!("{crate_name}-{pid:07}");
            let path = Path::new(out).join(filename);
            measureme::Profiler::new(path).expect("Couldn't create `measureme` profiler")
        });
        let rng = StdRng::seed_from_u64(config.seed.unwrap_or(0));
        let borrow_tracker = config.borrow_tracker.map(|bt| bt.instantiate_global_state(config));
        let data_race = if config.genmc_config.is_some() {
            // `genmc_ctx` persists across executions, so we don't create a new one here.
            GlobalDataRaceHandler::Genmc(genmc_ctx.unwrap())
        } else if config.data_race_detector {
            GlobalDataRaceHandler::Vclocks(Box::new(data_race::GlobalState::new(config)))
        } else {
            GlobalDataRaceHandler::None
        };
        // Determine page size, stack address, and stack size.
        // These values are mostly meaningless, but the stack address is also where we start
        // allocating physical integer addresses for all allocations.
        let page_size = if let Some(page_size) = config.page_size {
            page_size
        } else {
            let target = &tcx.sess.target;
            match target.arch {
                Arch::Wasm32 | Arch::Wasm64 => 64 * 1024, // https://webassembly.github.io/spec/core/exec/runtime.html#memory-instances
                Arch::AArch64 if target.is_like_darwin => {
                    // No "definitive" source, but see:
                    // https://www.wwdcnotes.com/notes/wwdc20/10214/
                    // https://github.com/ziglang/zig/issues/11308 etc.
                    16 * 1024
                }
                _ => 4 * 1024,
            }
        };
        // On 16bit targets, 32 pages is more than the entire address space!
        let stack_addr = if tcx.pointer_size().bits() < 32 { page_size } else { page_size * 32 };
        let stack_size =
            if tcx.pointer_size().bits() < 32 { page_size * 4 } else { page_size * 16 };
        assert!(
            usize::try_from(config.num_cpus).unwrap() <= shims::cpu_affinity::MAX_CPUS,
            "miri only supports up to {} CPUs, but {} were configured",
            shims::cpu_affinity::MAX_CPUS,
            config.num_cpus
        );
        let threads = ThreadManager::new(config);
        let thread_cpu_affinity =
            if matches!(&tcx.sess.target.os, Os::Linux | Os::FreeBsd | Os::Android)
                && !is_no_core(tcx)
            {
                let mut affinity = FxHashMap::default();
                affinity.insert(
                    threads.active_thread(),
                    shims::CpuAffinityMask::new(&layout_cx, config.num_cpus),
                );
                Some(affinity)
            } else {
                None
            };
        let blocking_io = BlockingIoManager::new(config.isolated_op == IsolatedOp::Allow)
            .expect("Couldn't create poll instance");
        let alloc_addresses =
            RefCell::new(alloc_addresses::GlobalStateInner::new(config, stack_addr, tcx));

        MiriMachine {
            tcx,
            borrow_tracker,
            data_race,
            alloc_addresses,
            // `env_vars` depends on a full interpreter so we cannot properly initialize it yet.
            env_vars: EnvVars::default(),
            main_fn_ret_place: None,
            argc: None,
            argv: None,
            cmd_line: None,
            tls: TlsData::default(),
            isolated_op: config.isolated_op,
            validation: config.validation,
            fds: shims::FdTable::init(config.mute_stdout_stderr),
            delayed_readiness_updates: Rc::new(shims::DelayedReadinessUpdates::default()),
            dirs: Default::default(),
            layouts,
            threads,
            thread_cpu_affinity,
            blocking_io,
            static_roots: Vec::new(),
            profiler,
            string_cache: Default::default(),
            exported_symbols_cache: RefCell::new(FxHashMap::default()),
            backtrace_style: config.backtrace_style,
            user_relevant_crates,
            extern_statics: FxHashMap::default(),
            extern_statics_imports: FxHashMap::default(),
            extern_static_weak_import_default: None,
            rng: RefCell::new(rng),
            allocator: (!config.native_lib.is_empty())
                .then(|| Rc::new(RefCell::new(crate::alloc::isolated_alloc::IsolatedAlloc::new()))),
            tracked_alloc_ids: config.tracked_alloc_ids.clone(),
            track_alloc_accesses: config.track_alloc_accesses,
            check_alignment: config.check_alignment,
            cmpxchg_weak_failure_rate: config.cmpxchg_weak_failure_rate,
            preemption_rate: config.preemption_rate,
            report_progress: config.report_progress,
            basic_block_count: 0,
            monotonic_clock: MonotonicClock::new(config.isolated_op == IsolatedOp::Allow),
            #[cfg(all(feature = "native-lib", unix))]
            native_lib: config.native_lib.iter().map(|lib_file_path| {
                let host_triple = rustc_session::config::host_tuple();
                let target_triple = tcx.sess.opts.target_triple.tuple();
                // Check if host target == the session target.
                if host_triple != target_triple {
                    panic!(
                        "calling native C functions in linked .so file requires host and target to be the same: \
                        host={host_triple}, target={target_triple}",
                    );
                }
                // Note: it is the user's responsibility to provide a correct SO file.
                // WATCH OUT: If an invalid/incorrect SO file is specified, this can cause
                // undefined behaviour in Miri itself!
                (
                    unsafe {
                        libloading::Library::new(lib_file_path)
                            .expect("failed to read specified extern shared object file")
                    },
                    lib_file_path.clone(),
                )
            }).collect(),
            #[cfg(all(feature = "native-lib", unix))]
            native_lib_ecx_interchange: Box::leak(Box::new(Cell::new(0))),
            #[cfg(not(all(feature = "native-lib", unix)))]
            native_lib: config.native_lib.iter().map(|_| {
                panic!("calling functions from native libraries via FFI is not supported in this build of Miri")
            }).collect(),
            gc_interval: config.gc_interval,
            since_gc: 0,
            num_cpus: config.num_cpus,
            page_size,
            stack_addr,
            stack_size,
            collect_leak_backtraces: config.collect_leak_backtraces,
            allocation_spans: RefCell::new(FxHashMap::default()),
            symbolic_alignment: RefCell::new(FxHashMap::default()),
            union_data_ranges: FxHashMap::default(),
            pthread_mutex_sanity: Cell::new(false),
            pthread_rwlock_sanity: Cell::new(false),
            pthread_condvar_sanity: Cell::new(false),
            allocator_shim_symbols: Self::allocator_shim_symbols(tcx),
            mangle_internal_symbol_cache: Default::default(),
            float_nondet: config.float_nondet,
            float_rounding_error: config.float_rounding_error,
            short_fd_operations: config.short_fd_operations,
        }
    }

    fn allocator_shim_symbols(
        tcx: TyCtxt<'tcx>,
    ) -> FxHashMap<Symbol, Either<Symbol, SpecialAllocatorMethod>> {
        use rustc_codegen_ssa::base::allocator_shim_contents;

        // codegen uses `allocator_kind_for_codegen` here, but that's only needed to deal with
        // dylibs which we do not support.
        let Some(kind) = tcx.allocator_kind(()) else {
            return Default::default();
        };
        let methods = allocator_shim_contents(tcx, kind);
        let mut symbols = FxHashMap::default();
        for method in methods {
            let from_name = Symbol::intern(&mangle_internal_symbol(
                tcx,
                &allocator::global_fn_name(method.name),
            ));
            let to = match method.special {
                Some(special) => Either::Right(special),
                None =>
                    Either::Left(Symbol::intern(&mangle_internal_symbol(
                        tcx,
                        &allocator::default_fn_name(method.name),
                    ))),
            };
            symbols.try_insert(from_name, to).unwrap();
        }
        symbols
    }

    /// Retrieve the list of user-relevant crates based on MIRI_LOCAL_CRATES as set by cargo-miri,
    /// and extra crates set in the config.
    fn get_user_relevant_crates(tcx: TyCtxt<'_>, config: &MiriConfig) -> Vec<CrateNum> {
        // Convert the local crate names from the passed-in config into CrateNums so that they can
        // be looked up quickly during execution
        let local_crate_names = std::env::var("MIRI_LOCAL_CRATES")
            .map(|crates| crates.split(',').map(|krate| krate.to_string()).collect::<Vec<_>>())
            .unwrap_or_default();
        let mut local_crates = Vec::new();
        for &crate_num in tcx.crates(()) {
            let name = tcx.crate_name(crate_num);
            let name = name.as_str();
            if local_crate_names
                .iter()
                .chain(&config.user_relevant_crates)
                .any(|local_name| local_name == name)
            {
                local_crates.push(crate_num);
            }
        }
        local_crates
    }

    pub(crate) fn late_init(
        ecx: &mut MiriInterpCx<'tcx>,
        config: &MiriConfig,
        on_main_stack_empty: StackEmptyCallback<'tcx>,
    ) -> InterpResult<'tcx> {
        EnvVars::init(ecx, config)?;
        MiriMachine::init_extern_statics(ecx)?;
        ThreadManager::init(ecx, on_main_stack_empty);
        interp_ok(())
    }

    pub(crate) fn communicate(&self) -> bool {
        self.isolated_op == IsolatedOp::Allow
    }

    /// Check whether the stack frame that this `FrameInfo` refers to is part of a local crate.
    pub(crate) fn is_local(&self, instance: ty::Instance<'tcx>) -> bool {
        let def_id = instance.def_id();
        def_id.is_local() || self.user_relevant_crates.contains(&def_id.krate)
    }

    /// Called when the interpreter is going to shut down abnormally, such as due to a Ctrl-C.
    pub(crate) fn handle_abnormal_termination(&mut self) {
        // All strings in the profile data are stored in a single string table which is not
        // written to disk until the profiler is dropped. If the interpreter exits without dropping
        // the profiler, it is not possible to interpret the profile data and all measureme tools
        // will panic when given the file.
        drop(self.profiler.take());
    }

    pub(crate) fn page_align(&self) -> Align {
        Align::from_bytes(self.page_size).unwrap()
    }

    pub(crate) fn allocated_span(&self, alloc_id: AllocId) -> Option<SpanData> {
        self.allocation_spans
            .borrow()
            .get(&alloc_id)
            .map(|(allocated, _deallocated)| allocated.data())
    }

    pub(crate) fn deallocated_span(&self, alloc_id: AllocId) -> Option<SpanData> {
        self.allocation_spans
            .borrow()
            .get(&alloc_id)
            .and_then(|(_allocated, deallocated)| *deallocated)
            .map(Span::data)
    }

    fn init_allocation(
        ecx: &MiriInterpCx<'tcx>,
        id: AllocId,
        kind: MemoryKind,
        size: Size,
        align: Align,
    ) -> InterpResult<'tcx, AllocExtra<'tcx>> {
        if ecx.machine.tracked_alloc_ids.contains(&id) {
            ecx.emit_diagnostic(NonHaltingDiagnostic::TrackingAlloc(id, size, align));
        }

        let borrow_tracker = ecx
            .machine
            .borrow_tracker
            .as_ref()
            .map(|bt| bt.borrow_mut().new_allocation(id, size, kind, &ecx.machine));

        let data_race = match &ecx.machine.data_race {
            GlobalDataRaceHandler::None => AllocDataRaceHandler::None,
            GlobalDataRaceHandler::Vclocks(data_race) =>
                AllocDataRaceHandler::Vclocks(
                    data_race::AllocState::new_allocation(
                        data_race,
                        &ecx.machine.threads,
                        size,
                        kind,
                        ecx.machine.current_user_relevant_span(),
                    ),
                    data_race.weak_memory.then(weak_memory::AllocState::new_allocation),
                ),
            GlobalDataRaceHandler::Genmc(_genmc_ctx) => {
                // GenMC learns about new allocations directly from the alloc_addresses module,
                // since it has to be able to control the address at which they are placed.
                AllocDataRaceHandler::Genmc
            }
        };

        // If an allocation is leaked, we want to report a backtrace to indicate where it was
        // allocated. We don't need to record a backtrace for allocations which are allowed to
        // leak.
        let backtrace = if kind.may_leak() || !ecx.machine.collect_leak_backtraces {
            None
        } else {
            Some(ecx.generate_stacktrace())
        };

        if matches!(kind, MemoryKind::Machine(kind) if kind.should_save_allocation_span()) {
            ecx.machine
                .allocation_spans
                .borrow_mut()
                .insert(id, (ecx.machine.current_user_relevant_span(), None));
        }

        interp_ok(AllocExtra {
            borrow_tracker,
            data_race,
            backtrace,
            sync_objs: BTreeMap::default(),
        })
    }
}

impl VisitProvenance for MiriMachine<'_> {
    fn visit_provenance(&self, visit: &mut VisitWith<'_>) {
        #[rustfmt::skip]
        let MiriMachine {
            threads,
            thread_cpu_affinity: _,
            tls,
            env_vars,
            main_fn_ret_place,
            argc,
            argv,
            cmd_line,
            extern_statics,
            extern_statics_imports,
            extern_static_weak_import_default,
            dirs,
            borrow_tracker,
            data_race,
            alloc_addresses,
            fds,
            blocking_io:_,
            delayed_readiness_updates: _,
            tcx: _,
            isolated_op: _,
            validation: _,
            monotonic_clock: _,
            layouts: _,
            static_roots: _,
            profiler: _,
            string_cache: _,
            exported_symbols_cache: _,
            backtrace_style: _,
            user_relevant_crates: _,
            rng: _,
            allocator: _,
            tracked_alloc_ids: _,
            track_alloc_accesses: _,
            check_alignment: _,
            cmpxchg_weak_failure_rate: _,
            preemption_rate: _,
            report_progress: _,
            basic_block_count: _,
            native_lib: _,
            #[cfg(all(feature = "native-lib", unix))]
            native_lib_ecx_interchange: _,
            gc_interval: _,
            since_gc: _,
            num_cpus: _,
            page_size: _,
            stack_addr: _,
            stack_size: _,
            collect_leak_backtraces: _,
            allocation_spans: _,
            symbolic_alignment: _,
            union_data_ranges: _,
            pthread_mutex_sanity: _,
            pthread_rwlock_sanity: _,
            pthread_condvar_sanity: _,
            allocator_shim_symbols: _,
            mangle_internal_symbol_cache: _,
            float_nondet: _,
            float_rounding_error: _,
            short_fd_operations: _,
        } = self;

        threads.visit_provenance(visit);
        tls.visit_provenance(visit);
        env_vars.visit_provenance(visit);
        dirs.visit_provenance(visit);
        fds.visit_provenance(visit);
        data_race.visit_provenance(visit);
        borrow_tracker.visit_provenance(visit);
        alloc_addresses.visit_provenance(visit);
        main_fn_ret_place.visit_provenance(visit);
        argc.visit_provenance(visit);
        argv.visit_provenance(visit);
        cmd_line.visit_provenance(visit);
        extern_static_weak_import_default.visit_provenance(visit);
        extern_statics.visit_provenance(visit);
        extern_statics_imports.visit_provenance(visit);
    }
}

/// A rustc InterpCx for Miri.
pub type MiriInterpCx<'tcx> = InterpCx<'tcx, MiriMachine<'tcx>>;

/// A little trait that's useful to be inherited by extension traits.
pub trait MiriInterpCxExt<'tcx> {
    fn eval_context_ref<'a>(&'a self) -> &'a MiriInterpCx<'tcx>;
    fn eval_context_mut<'a>(&'a mut self) -> &'a mut MiriInterpCx<'tcx>;
}
impl<'tcx> MiriInterpCxExt<'tcx> for MiriInterpCx<'tcx> {
    #[inline(always)]
    fn eval_context_ref(&self) -> &MiriInterpCx<'tcx> {
        self
    }
    #[inline(always)]
    fn eval_context_mut(&mut self) -> &mut MiriInterpCx<'tcx> {
        self
    }
}

/// Machine hook implementations.
impl<'tcx> Machine<'tcx> for MiriMachine<'tcx> {
    type MemoryKind = MiriMemoryKind;
    type ExtraFnVal = DynSym;

    type FrameExtra = FrameExtra<'tcx>;
    type AllocExtra = AllocExtra<'tcx>;

    type Provenance = Provenance;
    type ProvenanceExtra = ProvenanceExtra;
    type Bytes = MiriAllocBytes;

    type MemoryMap =
        MonoHashMap<AllocId, (MemoryKind, Allocation<Provenance, Self::AllocExtra, Self::Bytes>)>;

    const GLOBAL_KIND: Option<MiriMemoryKind> = Some(MiriMemoryKind::Global);

    const PANIC_ON_ALLOC_FAIL: bool = false;

    #[inline(always)]
    fn enforce_alignment(ecx: &MiriInterpCx<'tcx>) -> bool {
        ecx.machine.check_alignment != AlignmentCheck::None
    }

    #[inline(always)]
    fn alignment_check(
        ecx: &MiriInterpCx<'tcx>,
        alloc_id: AllocId,
        alloc_align: Align,
        alloc_kind: AllocKind,
        offset: Size,
        align: Align,
    ) -> Option<Misalignment> {
        if ecx.machine.check_alignment != AlignmentCheck::Symbolic {
            // Just use the built-in check.
            return None;
        }
        if alloc_kind != AllocKind::LiveData {
            // Can't have any extra info here.
            return None;
        }
        // Let's see which alignment we have been promised for this allocation.
        let (promised_offset, promised_align) = ecx
            .machine
            .symbolic_alignment
            .borrow()
            .get(&alloc_id)
            .copied()
            .unwrap_or((Size::ZERO, alloc_align));
        if promised_align < align {
            // Definitely not enough.
            Some(Misalignment { has: promised_align, required: align })
        } else {
            // What's the offset between us and the promised alignment?
            let distance = offset.bytes().wrapping_sub(promised_offset.bytes());
            // That must also be aligned.
            if distance.is_multiple_of(align.bytes()) {
                // All looking good!
                None
            } else {
                // The biggest power of two through which `distance` is divisible.
                let distance_pow2 = 1 << distance.trailing_zeros();
                Some(Misalignment {
                    has: Align::from_bytes(distance_pow2).unwrap(),
                    required: align,
                })
            }
        }
    }

    #[inline(always)]
    fn enforce_validity(ecx: &MiriInterpCx<'tcx>, _layout: TyAndLayout<'tcx>) -> bool {
        ecx.machine.validation != ValidationMode::No
    }
    #[inline(always)]
    fn enforce_validity_recursively(
        ecx: &InterpCx<'tcx, Self>,
        _layout: TyAndLayout<'tcx>,
    ) -> bool {
        ecx.machine.validation == ValidationMode::Deep
    }

    #[inline(always)]
    fn ignore_optional_overflow_checks(ecx: &MiriInterpCx<'tcx>) -> bool {
        !ecx.tcx.sess.overflow_checks()
    }

    fn check_fn_target_features(
        ecx: &MiriInterpCx<'tcx>,
        instance: ty::Instance<'tcx>,
    ) -> InterpResult<'tcx> {
        let attrs = ecx.tcx.codegen_instance_attrs(instance.def);
        if attrs
            .target_features
            .iter()
            .any(|feature| !ecx.tcx.sess.internal_target_features.contains(&feature.name))
        {
            let unavailable = attrs
                .target_features
                .iter()
                .filter(|&feature| {
                    feature.kind != TargetFeatureKind::Implied
                        && !ecx.tcx.sess.internal_target_features.contains(&feature.name)
                })
                .fold(String::new(), |mut s, feature| {
                    if !s.is_empty() {
                        s.push_str(", ");
                    }
                    s.push_str(feature.name.as_str());
                    s
                });
            let msg = format!(
                "calling a function that requires unavailable target features: {unavailable}"
            );
            // On WASM, this is not UB, but instead gets rejected during validation of the module
            // (see #84988).
            if ecx.tcx.sess.target.is_like_wasm {
                throw_machine_stop!(TerminationInfo::Abort(msg));
            } else {
                throw_ub_format!("{msg}");
            }
        }
        interp_ok(())
    }

    #[inline(always)]
    fn find_mir_or_eval_fn(
        ecx: &mut MiriInterpCx<'tcx>,
        instance: ty::Instance<'tcx>,
        abi: &FnAbi<'tcx, Ty<'tcx>>,
        args: &[FnArg<'tcx>],
        dest: &PlaceTy<'tcx>,
        ret: Option<mir::BasicBlock>,
        unwind: mir::UnwindAction,
    ) -> InterpResult<'tcx, Option<(&'tcx mir::Body<'tcx>, ty::Instance<'tcx>)>> {
        // For foreign items, try to see if we can emulate them.
        if ecx.tcx.is_foreign_item(instance.def_id()) {
            let _trace = enter_trace_span!("emulate_foreign_item");
            // An external function call that does not have a MIR body. We either find MIR elsewhere
            // or emulate its effect.
            // This will be Ok(None) if we're emulating the intrinsic entirely within Miri (no need
            // to run extra MIR), and Ok(Some(body)) if we found MIR to run for the
            // foreign function
            // Any needed call to `goto_block` will be performed by `emulate_foreign_item`.
            let args = MiriInterpCx::copy_fn_args(args); // FIXME: Should `InPlace` arguments be reset to uninit?
            let link_name = Symbol::intern(ecx.tcx.symbol_name(instance).name);
            return ecx.emulate_foreign_item(link_name, abi, &args, dest, ret, unwind);
        }

        if ecx.machine.data_race.as_genmc_ref().is_some()
            && ecx.genmc_intercept_function(instance, args, dest)?
        {
            ecx.return_to_block(ret)?;
            return interp_ok(None);
        }

        // Otherwise, load the MIR.
        let _trace = enter_trace_span!("load_mir");
        interp_ok(Some((ecx.load_mir(instance.def, None)?, instance)))
    }

    #[inline(always)]
    fn call_extra_fn(
        ecx: &mut MiriInterpCx<'tcx>,
        fn_val: DynSym,
        abi: &FnAbi<'tcx, Ty<'tcx>>,
        args: &[FnArg<'tcx>],
        dest: &PlaceTy<'tcx>,
        ret: Option<mir::BasicBlock>,
        unwind: mir::UnwindAction,
    ) -> InterpResult<'tcx> {
        let args = MiriInterpCx::copy_fn_args(args); // FIXME: Should `InPlace` arguments be reset to uninit?
        ecx.emulate_dyn_sym(fn_val, abi, &args, dest, ret, unwind)
    }

    #[inline(always)]
    fn call_intrinsic(
        ecx: &mut MiriInterpCx<'tcx>,
        instance: ty::Instance<'tcx>,
        args: &[OpTy<'tcx>],
        dest: &PlaceTy<'tcx>,
        ret: Option<mir::BasicBlock>,
        unwind: mir::UnwindAction,
    ) -> InterpResult<'tcx, Option<ty::Instance<'tcx>>> {
        ecx.call_intrinsic(instance, args, dest, ret, unwind)
    }

    #[inline(always)]
    fn call_llvm_intrinsic(
        ecx: &mut MiriInterpCx<'tcx>,
        instance: ty::Instance<'tcx>,
        args: &[OpTy<'tcx>],
        dest: &PlaceTy<'tcx>,
        ret: Option<mir::BasicBlock>,
    ) -> InterpResult<'tcx, ()> {
        ecx.call_llvm_intrinsic(instance, args, dest, ret)
    }

    #[inline(always)]
    fn assert_panic(
        ecx: &mut MiriInterpCx<'tcx>,
        msg: &mir::AssertMessage<'tcx>,
        unwind: mir::UnwindAction,
    ) -> InterpResult<'tcx> {
        ecx.assert_panic(msg, unwind)
    }

    fn panic_nounwind(ecx: &mut InterpCx<'tcx, Self>, msg: &str) -> InterpResult<'tcx> {
        ecx.start_panic_nounwind(msg)
    }

    fn unwind_terminate(
        ecx: &mut InterpCx<'tcx, Self>,
        reason: mir::UnwindTerminateReason,
    ) -> InterpResult<'tcx> {
        // Call the lang item.
        let panic = ecx.tcx.lang_items().get(reason.lang_item()).unwrap();
        let panic = ty::Instance::mono(ecx.tcx.tcx, panic);
        ecx.call_function(
            panic,
            ExternAbi::Rust,
            &[],
            None,
            ReturnContinuation::Goto { ret: None, unwind: mir::UnwindAction::Unreachable },
        )?;
        interp_ok(())
    }

    #[inline(always)]
    fn binary_ptr_op(
        ecx: &MiriInterpCx<'tcx>,
        bin_op: mir::BinOp,
        left: &ImmTy<'tcx>,
        right: &ImmTy<'tcx>,
    ) -> InterpResult<'tcx, ImmTy<'tcx>> {
        ecx.binary_ptr_op(bin_op, left, right)
    }

    fn atomic_load(
        ecx: &MiriInterpCx<'tcx>,
        place: &MPlaceTy<'tcx>,
        ordering: AtomicOrdering,
    ) -> InterpResult<'tcx, Scalar> {
        ecx.read_scalar_atomic(place, AtomicReadOrd::from(ordering))
    }

    fn atomic_store(
        ecx: &mut MiriInterpCx<'tcx>,
        place: &MPlaceTy<'tcx>,
        val: &ImmTy<'tcx>,
        ordering: AtomicOrdering,
    ) -> InterpResult<'tcx> {
        ecx.write_scalar_atomic(val.to_scalar(), place, AtomicWriteOrd::from(ordering))
    }

    fn atomic_rmw(
        ecx: &mut MiriInterpCx<'tcx>,
        place: &MPlaceTy<'tcx>,
        op: AtomicRmwOp,
        operand: &ImmTy<'tcx>,
        ordering: AtomicOrdering,
    ) -> InterpResult<'tcx, Scalar> {
        ecx.atomic_rmw(place, operand, op, AtomicRwOrd::from(ordering))
    }

    fn atomic_compare_exchange(
        ecx: &mut MiriInterpCx<'tcx>,
        place: &MPlaceTy<'tcx>,
        expected_old: &ImmTy<'tcx>,
        new: &ImmTy<'tcx>,
        can_fail_spuriously: bool,
        success_ordering: AtomicOrdering,
        failure_ordering: AtomicOrdering,
    ) -> InterpResult<'tcx, (Scalar, bool)> {
        ecx.atomic_compare_exchange(
            place,
            expected_old,
            new.to_scalar(),
            AtomicRwOrd::from(success_ordering),
            AtomicReadOrd::from(failure_ordering),
            can_fail_spuriously,
        )
    }

    fn atomic_fence(
        ecx: &MiriInterpCx<'tcx>,
        ordering: AtomicOrdering,
        singlethread: bool,
    ) -> InterpResult<'tcx> {
        if singlethread {
            // We don't support signal handlers or interrupts so this is a NOP.
            return interp_ok(());
        }
        ecx.atomic_fence(AtomicFenceOrd::from(ordering))
    }

    #[inline(always)]
    fn generate_nan<F1: Float + FloatConvert<F2>, F2: Float>(
        ecx: &InterpCx<'tcx, Self>,
        inputs: &[F1],
    ) -> F2 {
        ecx.generate_nan(inputs)
    }

    #[inline(always)]
    fn apply_float_nondet(
        ecx: &mut InterpCx<'tcx, Self>,
        val: ImmTy<'tcx>,
    ) -> InterpResult<'tcx, ImmTy<'tcx>> {
        crate::math::apply_random_float_error_to_imm(ecx, val, 4)
    }

    #[inline(always)]
    fn equal_float_min_max<F: Float>(ecx: &MiriInterpCx<'tcx>, a: F, b: F) -> F {
        ecx.equal_float_min_max(a, b)
    }

    #[inline(always)]
    fn float_fuse_mul_add(ecx: &InterpCx<'tcx, Self>) -> bool {
        ecx.machine.float_nondet && ecx.machine.rng.borrow_mut().random()
    }

    #[inline(always)]
    fn runtime_checks(
        ecx: &InterpCx<'tcx, Self>,
        r: mir::RuntimeChecks,
    ) -> InterpResult<'tcx, bool> {
        interp_ok(r.value(ecx.tcx.sess))
    }

    #[inline(always)]
    fn thread_local_static_pointer(
        ecx: &mut MiriInterpCx<'tcx>,
        def_id: DefId,
    ) -> InterpResult<'tcx, StrictPointer> {
        ecx.get_or_create_thread_local_alloc(def_id)
    }

    fn extern_static_pointer(
        ecx: &MiriInterpCx<'tcx>,
        def_id: DefId,
    ) -> InterpResult<'tcx, StrictPointer> {
        let link_name = Symbol::intern(ecx.tcx.symbol_name(Instance::mono(*ecx.tcx, def_id)).name);
        let def_ty = ecx.tcx.type_of(def_id).instantiate_identity().skip_norm_wip();
        let extern_decl_layout =
            ecx.tcx.layout_of(ecx.typing_env().as_query_input(def_ty)).unwrap();

        // Look up the `ptr` in the right map, depending on whether this is an "import"
        // static or a real one.
        let ptr = match ecx.tcx.codegen_fn_attrs(def_id).import_linkage {
            None => ecx.machine.extern_statics.get(&link_name),
            Some(_) => ecx.machine.extern_statics_imports.get(&link_name),
        };
        if let Some(&ptr) = ptr {
            ecx.check_shim_symbol_clash(link_name)?;
            // Various parts of the engine rely on `get_alloc_info` for size and alignment
            // information. That uses the type information of this static.
            // Make sure it matches the Miri allocation for this.
            let Provenance::Concrete { alloc_id, .. } = ptr.provenance else {
                panic!("extern_statics cannot contain wildcards")
            };
            let info = ecx.get_alloc_info(alloc_id);
            if extern_decl_layout.size > info.size || extern_decl_layout.align.abi > info.align {
                throw_ub_format!(
                    "extern static `{link_name}` has been declared as `{krate}::{name}` \
                    with a size of {decl_size} bytes and alignment of {decl_align} bytes, \
                    but Miri emulates it via an extern static shim \
                    with a size of {shim_size} bytes and alignment of {shim_align} bytes",
                    name = ecx.tcx.def_path_str(def_id),
                    krate = ecx.tcx.crate_name(def_id.krate),
                    decl_size = extern_decl_layout.size.bytes(),
                    decl_align = extern_decl_layout.align.bytes(),
                    shim_size = info.size.bytes(),
                    shim_align = info.align.bytes(),
                )
            }
            interp_ok(ptr)
        } else if ecx.tcx.codegen_fn_attrs(def_id).import_linkage == Some(Linkage::ExternalWeak) {
            // Symbols with weak linkage default to null if they are not defined. However we can't
            // create new allocations here. On the plus side we know rustc rejects non-ptr-sized
            // weak statics so we can just use a single global "null" allocation for all of them.
            // The memory we are assigning this address to is anyway somewhat "fake", it's an
            // indirection introduced by how Rust represents external symbols with linkage (see
            // <https://github.com/rust-lang/rust/issues/156468>). So we can just specify that such
            // memory does not have unique addresses, despite being technically a `static`.
            assert_eq!(
                extern_decl_layout.size,
                ecx.tcx.data_layout.pointer_size(),
                "non-pointer-sized weak static"
            );
            interp_ok(
                ecx.machine
                    .extern_static_weak_import_default
                    .expect("`missing_weak_symbol` should have been initialized"),
            )
        } else {
            // Look for a Rust static with this symbol name in the crate graph.
            let Some(instance) = ecx.lookup_exported_static(link_name)? else {
                throw_unsup_format!("extern static `{link_name}` is not supported by Miri");
            };
            // Evaluate the static to get its allocation.
            let place = ecx.eval_global(instance)?;
            let static_ptr = place.ptr().into_pointer_or_addr().unwrap();
            // Validate the allocation matches the declared size and alignment.
            let alloc_id = static_ptr.provenance.get_alloc_id().unwrap();
            let info = ecx.get_alloc_info(alloc_id);
            if extern_decl_layout.size > info.size || extern_decl_layout.align.abi > info.align {
                throw_ub_format!(
                    "extern static `{link_name}` has been declared as `{krate}::{name}` \
                    with a size of {decl_size} bytes and alignment of {decl_align} bytes, \
                    but the exported static with that name has a size of {shim_size} bytes and \
                    alignment of {shim_align} bytes",
                    name = ecx.tcx.def_path_str(def_id),
                    krate = ecx.tcx.crate_name(def_id.krate),
                    decl_size = extern_decl_layout.size.bytes(),
                    decl_align = extern_decl_layout.align.bytes(),
                    shim_size = info.size.bytes(),
                    shim_align = info.align.bytes(),
                )
            }
            // Check that the mutability of the declared static matches that of the backing.
            // If the backing static can be modified (because it is a `static mut`, or because
            // it is a `static` whose type has interior mutability) while the declaration here
            // is a non-mut `static` with a `Freeze` type, then the compiler's assumption that
            // the value never changes may be violated, so this may cause UB.
            // This is somehow defensive, as the allocation might be mutable but no mutation
            // ever happens, but this is probably the most precise thing we can do.
            // Specially, the second case is very defensive and we may be able to lift it.
            let DefKind::Static { mutability, .. } = ecx.tcx.def_kind(def_id) else {
                unreachable!("`{def_id:?}` is not a static");
            };
            let decl_is_mut =
                !(mutability == Mutability::Not && ecx.type_is_freeze(extern_decl_layout.ty));
            let backing_is_mut = ecx.get_alloc_mutability(alloc_id)? == Mutability::Mut;
            if !decl_is_mut && backing_is_mut {
                throw_ub_format!(
                    "extern static `{krate}::{name}` is declared as an immutable `static`, \
                    but the backing static is mutable",
                    name = ecx.tcx.def_path_str(def_id),
                    krate = ecx.tcx.crate_name(def_id.krate),
                )
            }
            if decl_is_mut && !backing_is_mut {
                throw_ub_format!(
                    "extern static `{krate}::{name}` is declared as an mutable `static`, \
                    but the backing static is immutable",
                    name = ecx.tcx.def_path_str(def_id),
                    krate = ecx.tcx.crate_name(def_id.krate),
                )
            }
            interp_ok(static_ptr)
        }
    }

    fn init_local_allocation(
        ecx: &MiriInterpCx<'tcx>,
        id: AllocId,
        kind: MemoryKind,
        size: Size,
        align: Align,
    ) -> InterpResult<'tcx, Self::AllocExtra> {
        assert!(kind != MiriMemoryKind::Global.into());
        MiriMachine::init_allocation(ecx, id, kind, size, align)
    }

    fn adjust_alloc_root_pointer(
        ecx: &MiriInterpCx<'tcx>,
        ptr: interpret::Pointer<CtfeProvenance>,
        kind: Option<MemoryKind>,
    ) -> InterpResult<'tcx, interpret::Pointer<Provenance>> {
        let kind = kind.expect("we set our GLOBAL_KIND so this cannot be None");
        let alloc_id = ptr.provenance.alloc_id();
        if cfg!(debug_assertions) {
            // The machine promises to never call us on thread-local or extern statics.
            match ecx.tcx.try_get_global_alloc(alloc_id) {
                Some(GlobalAlloc::Static(def_id)) if ecx.tcx.is_thread_local_static(def_id) => {
                    panic!("adjust_alloc_root_pointer called on thread-local static")
                }
                Some(GlobalAlloc::Static(def_id)) if ecx.tcx.is_foreign_item(def_id) => {
                    panic!("adjust_alloc_root_pointer called on extern static")
                }
                _ => {}
            }
        }
        // FIXME: can we somehow preserve the immutability of `ptr`?
        let tag = if let Some(borrow_tracker) = &ecx.machine.borrow_tracker {
            borrow_tracker.borrow_mut().root_ptr_tag(alloc_id, &ecx.machine)
        } else {
            // Value does not matter, SB is disabled
            BorTag::default()
        };
        ecx.adjust_alloc_root_pointer(ptr, tag, kind)
    }

    /// Called on `usize as ptr` casts.
    #[inline(always)]
    fn ptr_from_addr_cast(ecx: &MiriInterpCx<'tcx>, addr: u64) -> InterpResult<'tcx, Pointer> {
        ecx.ptr_from_addr_cast(addr)
    }

    /// Called on `ptr as usize` casts.
    /// (Actually computing the resulting `usize` doesn't need machine help,
    /// that's just `Scalar::try_to_int`.)
    #[inline(always)]
    fn expose_provenance(
        ecx: &InterpCx<'tcx, Self>,
        provenance: Self::Provenance,
    ) -> InterpResult<'tcx> {
        ecx.expose_provenance(provenance)
    }

    /// Convert a pointer with provenance into an allocation-offset pair and extra provenance info.
    /// `size` says how many bytes of memory are expected at that pointer. The *sign* of `size` can
    /// be used to disambiguate situations where a wildcard pointer sits right in between two
    /// allocations.
    ///
    /// If `ptr.provenance.get_alloc_id()` is `Some(p)`, the returned `AllocId` must be `p`.
    /// The resulting `AllocId` will just be used for that one step and the forgotten again
    /// (i.e., we'll never turn the data returned here back into a `Pointer` that might be
    /// stored in machine state).
    ///
    /// When this fails, that means the pointer does not point to a live allocation.
    fn ptr_get_alloc(
        ecx: &MiriInterpCx<'tcx>,
        ptr: StrictPointer,
        size: i64,
    ) -> Option<(AllocId, Size, Self::ProvenanceExtra)> {
        let rel = ecx.ptr_get_alloc(ptr, size);

        rel.map(|(alloc_id, size)| {
            let tag = match ptr.provenance {
                Provenance::Concrete { tag, .. } => ProvenanceExtra::Concrete(tag),
                Provenance::Wildcard => ProvenanceExtra::Wildcard,
            };
            (alloc_id, size, tag)
        })
    }

    /// Called to adjust global allocations to the Provenance and AllocExtra of this machine.
    ///
    /// If `alloc` contains pointers, then they are all pointing to globals.
    ///
    /// This should avoid copying if no work has to be done! If this returns an owned
    /// allocation (because a copy had to be done to adjust things), machine memory will
    /// cache the result. (This relies on `AllocMap::get_or` being able to add the
    /// owned allocation to the map even when the map is shared.)
    fn adjust_global_allocation<'b>(
        ecx: &InterpCx<'tcx, Self>,
        id: AllocId,
        alloc: &'b Allocation,
    ) -> InterpResult<'tcx, Cow<'b, Allocation<Self::Provenance, Self::AllocExtra, Self::Bytes>>>
    {
        let alloc = alloc.adjust_from_tcx(
            &ecx.tcx,
            |bytes, align| ecx.get_global_alloc_bytes(id, bytes, align),
            |ptr| ecx.global_root_pointer(ptr),
        )?;
        let kind = MiriMemoryKind::Global.into();
        let extra = MiriMachine::init_allocation(ecx, id, kind, alloc.size(), alloc.align)?;
        interp_ok(Cow::Owned(alloc.with_extra(extra)))
    }

    #[inline(always)]
    fn before_memory_read(
        _tcx: TyCtxtAt<'tcx>,
        machine: &Self,
        alloc_extra: &AllocExtra<'tcx>,
        ptr: Pointer,
        (alloc_id, prov_extra): (AllocId, Self::ProvenanceExtra),
        range: AllocRange,
    ) -> InterpResult<'tcx> {
        if machine.track_alloc_accesses && machine.tracked_alloc_ids.contains(&alloc_id) {
            machine.emit_diagnostic(NonHaltingDiagnostic::AccessedAlloc(
                alloc_id,
                range,
                borrow_tracker::AccessKind::Read,
            ));
        }
        // The order of checks is deliberate, to prefer reporting a data race over a borrow tracker error.
        match &machine.data_race {
            GlobalDataRaceHandler::None => {}
            GlobalDataRaceHandler::Genmc(genmc_ctx) =>
                genmc_ctx.memory_load(machine, ptr.addr(), range.size)?,
            GlobalDataRaceHandler::Vclocks(_data_race) => {
                let _trace = enter_trace_span!(data_race::before_memory_read);
                let AllocDataRaceHandler::Vclocks(data_race, _weak_memory) = &alloc_extra.data_race
                else {
                    unreachable!();
                };
                data_race.read_non_atomic(alloc_id, range, NaReadType::Read, None, machine)?;
            }
        }
        if let Some(borrow_tracker) = &alloc_extra.borrow_tracker {
            borrow_tracker.before_memory_read(alloc_id, prov_extra, range, machine)?;
        }
        // Check if there are any sync objects that would like to prevent reading this memory.
        for (_offset, obj) in alloc_extra.sync_objs.range(range.start..range.end()) {
            obj.on_access(concurrency::sync::AccessKind::Read)?;
        }

        interp_ok(())
    }

    #[inline(always)]
    fn before_memory_write(
        _tcx: TyCtxtAt<'tcx>,
        machine: &mut Self,
        alloc_extra: &mut AllocExtra<'tcx>,
        ptr: Pointer,
        (alloc_id, prov_extra): (AllocId, Self::ProvenanceExtra),
        range: AllocRange,
    ) -> InterpResult<'tcx> {
        if machine.track_alloc_accesses && machine.tracked_alloc_ids.contains(&alloc_id) {
            machine.emit_diagnostic(NonHaltingDiagnostic::AccessedAlloc(
                alloc_id,
                range,
                borrow_tracker::AccessKind::Write,
            ));
        }
        match &machine.data_race {
            GlobalDataRaceHandler::None => {}
            GlobalDataRaceHandler::Genmc(genmc_ctx) =>
                genmc_ctx.memory_store(machine, ptr.addr(), range.size)?,
            GlobalDataRaceHandler::Vclocks(_global_state) => {
                let _trace = enter_trace_span!(data_race::before_memory_write);
                let AllocDataRaceHandler::Vclocks(data_race, weak_memory) =
                    &mut alloc_extra.data_race
                else {
                    unreachable!()
                };
                data_race.write_non_atomic(alloc_id, range, NaWriteType::Write, None, machine)?;
                if let Some(weak_memory) = weak_memory {
                    weak_memory
                        .non_atomic_write(range, machine.data_race.as_vclocks_ref().unwrap());
                }
            }
        }
        if let Some(borrow_tracker) = &mut alloc_extra.borrow_tracker {
            borrow_tracker.before_memory_write(alloc_id, prov_extra, range, machine)?;
        }
        // Delete sync objects that don't like writes.
        // Most of the time, we can just skip this.
        if !alloc_extra.sync_objs.is_empty() {
            let mut to_delete = vec![];
            for (offset, obj) in alloc_extra.sync_objs.range(range.start..range.end()) {
                obj.on_access(concurrency::sync::AccessKind::Write)?;
                if obj.delete_on_write() {
                    to_delete.push(*offset);
                }
            }
            for offset in to_delete {
                alloc_extra.sync_objs.remove(&offset);
            }
        }
        interp_ok(())
    }

    #[inline(always)]
    fn before_memory_deallocation(
        _tcx: TyCtxtAt<'tcx>,
        machine: &mut Self,
        alloc_extra: &mut AllocExtra<'tcx>,
        ptr: Pointer,
        (alloc_id, prove_extra): (AllocId, Self::ProvenanceExtra),
        size: Size,
        align: Align,
        kind: MemoryKind,
    ) -> InterpResult<'tcx> {
        if machine.tracked_alloc_ids.contains(&alloc_id) {
            machine.emit_diagnostic(NonHaltingDiagnostic::FreedAlloc(alloc_id));
        }
        match &machine.data_race {
            GlobalDataRaceHandler::None => {}
            GlobalDataRaceHandler::Genmc(genmc_ctx) =>
                genmc_ctx.handle_dealloc(machine, alloc_id, ptr.addr(), kind)?,
            GlobalDataRaceHandler::Vclocks(_global_state) => {
                let _trace = enter_trace_span!(data_race::before_memory_deallocation);
                let data_race = alloc_extra.data_race.as_vclocks_mut().unwrap();
                data_race.write_non_atomic(
                    alloc_id,
                    alloc_range(Size::ZERO, size),
                    NaWriteType::Deallocate,
                    None,
                    machine,
                )?;
            }
        }
        if let Some(borrow_tracker) = &mut alloc_extra.borrow_tracker {
            borrow_tracker.before_memory_deallocation(alloc_id, prove_extra, size, machine)?;
        }
        // Check if there are any sync objects that would like to prevent freeing this memory.
        for obj in alloc_extra.sync_objs.values() {
            obj.on_access(concurrency::sync::AccessKind::Dealloc)?;
        }

        if let Some((_, deallocated_at)) = machine.allocation_spans.borrow_mut().get_mut(&alloc_id)
        {
            *deallocated_at = Some(machine.current_user_relevant_span());
        }
        machine.free_alloc_id(alloc_id, size, align, kind);
        interp_ok(())
    }

    #[inline(always)]
    fn retag_ptr_value(
        ecx: &mut InterpCx<'tcx, Self>,
        val: &ImmTy<'tcx>,
        ty: Ty<'tcx>,
    ) -> InterpResult<'tcx, Option<ImmTy<'tcx>>> {
        if ecx.machine.borrow_tracker.is_some() {
            ecx.retag_ptr_value(val, ty)
        } else {
            interp_ok(None)
        }
    }

    #[inline(always)]
    fn with_retag_mode<T>(
        ecx: &mut InterpCx<'tcx, Self>,
        mode: RetagMode,
        f: impl FnOnce(&mut InterpCx<'tcx, Self>) -> InterpResult<'tcx, T>,
    ) -> InterpResult<'tcx, T> {
        if ecx.machine.borrow_tracker.is_some() { ecx.with_retag_mode(mode, f) } else { f(ecx) }
    }

    fn protect_in_place_function_argument(
        ecx: &mut InterpCx<'tcx, Self>,
        place: &MPlaceTy<'tcx>,
    ) -> InterpResult<'tcx> {
        // If we have a borrow tracker, we also have it set up protection so that all reads *and
        // writes* during this call are insta-UB.
        let protected_place = if ecx.machine.borrow_tracker.is_some() {
            ecx.protect_place(place)?
        } else {
            // No borrow tracker.
            place.clone()
        };
        // We do need to write `uninit` so that even after the call ends, the former contents of
        // this place cannot be observed any more. We do the write after retagging so that for
        // Tree Borrows, this is considered to activate the new tag.
        // Conveniently this also ensures that the place actually points to suitable memory.
        ecx.write_uninit(&protected_place)?;
        // Now we throw away the protected place, ensuring its tag is never used again.
        interp_ok(())
    }

    #[inline(always)]
    fn init_frame(
        ecx: &mut InterpCx<'tcx, Self>,
        frame: Frame<'tcx, Provenance>,
    ) -> InterpResult<'tcx, Frame<'tcx, Provenance, FrameExtra<'tcx>>> {
        // Start recording our event before doing anything else
        let timing = if let Some(profiler) = ecx.machine.profiler.as_ref() {
            let fn_name = frame.instance().to_string();
            let entry = ecx.machine.string_cache.entry(fn_name.clone());
            let name = entry.or_insert_with(|| profiler.alloc_string(&*fn_name));

            Some(profiler.start_recording_interval_event_detached(
                *name,
                measureme::EventId::from_label(*name),
                ecx.active_thread().to_u32(),
            ))
        } else {
            None
        };

        let borrow_tracker = ecx.machine.borrow_tracker.as_ref();

        let extra = FrameExtra {
            borrow_tracker: borrow_tracker.map(|bt| bt.borrow_mut().new_frame()),
            catch_unwind: None,
            timing,
            user_relevance: ecx.machine.user_relevance(&frame),
            data_race: ecx
                .machine
                .data_race
                .as_vclocks_ref()
                .map(|_| data_race::FrameState::default()),
        };

        interp_ok(frame.with_extra(extra))
    }

    fn stack<'a>(
        ecx: &'a InterpCx<'tcx, Self>,
    ) -> &'a [Frame<'tcx, Self::Provenance, Self::FrameExtra>] {
        ecx.active_thread_stack()
    }

    fn stack_mut<'a>(
        ecx: &'a mut InterpCx<'tcx, Self>,
    ) -> &'a mut Vec<Frame<'tcx, Self::Provenance, Self::FrameExtra>> {
        ecx.active_thread_stack_mut()
    }

    fn before_terminator(ecx: &mut InterpCx<'tcx, Self>) -> InterpResult<'tcx> {
        ecx.machine.basic_block_count += 1u64; // a u64 that is only incremented by 1 will "never" overflow
        ecx.machine.since_gc += 1;
        // Possibly report our progress. This will point at the terminator we are about to execute.
        if let Some(report_progress) = ecx.machine.report_progress {
            if ecx.machine.basic_block_count.is_multiple_of(u64::from(report_progress)) {
                ecx.emit_diagnostic(NonHaltingDiagnostic::ProgressReport {
                    block_count: ecx.machine.basic_block_count,
                });
            }
        }

        // Search for BorTags to find all live pointers, then remove all other tags from borrow
        // stacks. Also clean up dropped readiness watchers from the global readiness interest
        // table and closed source file descriptions in the blocking I/O manager.
        // When debug assertions are enabled, run the GC as often as possible so that any cases
        // where it mistakenly removes an important tag become visible.
        if ecx.machine.gc_interval > 0 && ecx.machine.since_gc >= ecx.machine.gc_interval {
            ecx.machine.since_gc = 0;
            ecx.run_provenance_gc();
            ecx.machine.blocking_io.run_gc();
        }

        // These are our preemption points.
        // (This will only take effect after the terminator has been executed.)
        ecx.maybe_preempt_active_thread();

        // Make sure some time passes.
        ecx.machine.monotonic_clock.tick();

        interp_ok(())
    }

    #[inline(always)]
    fn after_stack_push(ecx: &mut InterpCx<'tcx, Self>) -> InterpResult<'tcx> {
        if ecx.frame().extra.user_relevance >= ecx.active_thread_ref().current_user_relevance() {
            // We just pushed a frame that's at least as relevant as the so-far most relevant frame.
            // That means we are now the most relevant frame.
            let stack_len = ecx.active_thread_stack().len();
            ecx.active_thread_mut().set_top_user_relevant_frame(stack_len - 1);
        }
        interp_ok(())
    }

    fn before_stack_pop(ecx: &mut InterpCx<'tcx, Self>) -> InterpResult<'tcx> {
        let frame = ecx.frame();
        // We want this *before* the return value copy, because the return place itself is protected
        // until we do `on_stack_pop` here, and we need to un-protect it to copy the return value.
        if ecx.machine.borrow_tracker.is_some() {
            ecx.on_stack_pop(frame)?;
        }
        if ecx
            .active_thread_ref()
            .top_user_relevant_frame()
            .expect("there should always be a most relevant frame for a non-empty stack")
            == ecx.frame_idx()
        {
            // We are popping the most relevant frame. We have no clue what the next relevant frame
            // below that is, so we recompute that.
            // (If this ever becomes a bottleneck, we could have `push` store the previous
            // user-relevant frame and restore that here.)
            // We have to skip the frame that is just being popped.
            ecx.active_thread_mut().recompute_top_user_relevant_frame(/* skip */ 1);
        }
        // tracing-tree can automatically annotate scope changes, but it gets very confused by our
        // concurrency and what it prints is just plain wrong. So we print our own information
        // instead. (Cc https://github.com/rust-lang/miri/issues/2266)
        info!("Leaving {}", ecx.frame().instance());
        interp_ok(())
    }

    #[inline(always)]
    fn after_stack_pop(
        ecx: &mut InterpCx<'tcx, Self>,
        frame: Frame<'tcx, Provenance, FrameExtra<'tcx>>,
        unwinding: bool,
    ) -> InterpResult<'tcx, ReturnAction> {
        let res = {
            // Move `frame` into a sub-scope so we control when it will be dropped.
            let mut frame = frame;
            let timing = frame.extra.timing.take();
            let res = ecx.handle_stack_pop_unwind(frame.extra, unwinding);
            if let Some(profiler) = ecx.machine.profiler.as_ref() {
                profiler.finish_recording_interval_event(timing.unwrap());
            }
            res
        };
        // Needs to be done after dropping frame to show up on the right nesting level.
        // (Cc https://github.com/rust-lang/miri/issues/2266)
        if !ecx.active_thread_stack().is_empty() {
            info!("Continuing in {}", ecx.frame().instance());
        }
        res
    }

    fn after_local_read(ecx: &InterpCx<'tcx, Self>, local: mir::Local) -> InterpResult<'tcx> {
        if let Some(data_race) = &ecx.frame().extra.data_race {
            let _trace = enter_trace_span!(data_race::after_local_read);
            data_race.local_read(local, &ecx.machine);
        }
        interp_ok(())
    }

    fn after_local_write(
        ecx: &mut InterpCx<'tcx, Self>,
        local: mir::Local,
        storage_live: bool,
    ) -> InterpResult<'tcx> {
        if let Some(data_race) = &ecx.frame().extra.data_race {
            let _trace = enter_trace_span!(data_race::after_local_write);
            data_race.local_write(local, storage_live, &ecx.machine);
        }
        interp_ok(())
    }

    fn after_local_moved_to_memory(
        ecx: &mut InterpCx<'tcx, Self>,
        local: mir::Local,
        mplace: &MPlaceTy<'tcx>,
    ) -> InterpResult<'tcx> {
        let Some(Provenance::Concrete { alloc_id, .. }) = mplace.ptr().provenance else {
            panic!("after_local_allocated should only be called on fresh allocations");
        };
        // Record the span where this was allocated: the declaration of the local.
        let local_decl = &ecx.frame().body().local_decls[local];
        let span = local_decl.source_info.span;
        ecx.machine.allocation_spans.borrow_mut().insert(alloc_id, (span, None));
        // The data race system has to fix the clocks used for this write.
        let (alloc_info, machine) = ecx.get_alloc_extra_mut(alloc_id)?;
        if let Some(data_race) =
            &machine.threads.active_thread_stack().last().unwrap().extra.data_race
        {
            let _trace = enter_trace_span!(data_race::after_local_moved_to_memory);
            data_race.local_moved_to_memory(
                local,
                alloc_info.data_race.as_vclocks_mut().unwrap(),
                machine,
            );
        }
        interp_ok(())
    }

    fn get_global_alloc_salt(
        ecx: &InterpCx<'tcx, Self>,
        instance: Option<ty::Instance<'tcx>>,
    ) -> usize {
        let unique = if let Some(instance) = instance {
            // Functions cannot be identified by pointers, as asm-equal functions can get
            // deduplicated by the linker (we set the "unnamed_addr" attribute for LLVM) and
            // functions can be duplicated across crates. We thus generate a new `AllocId` for every
            // mention of a function. This means that `main as fn() == main as fn()` is false, while
            // `let x = main as fn(); x == x` is true. However, as a quality-of-life feature it can
            // be useful to identify certain functions uniquely, e.g. for backtraces. So we identify
            // whether codegen will actually emit duplicate functions. It does that when they have
            // non-lifetime generics, or when they can be inlined. All other functions are given a
            // unique address. This is not a stable guarantee! The `inline` attribute is a hint and
            // cannot be relied upon for anything. But if we don't do this, the
            // `__rust_begin_short_backtrace`/`__rust_end_short_backtrace` logic breaks and panic
            // backtraces look terrible.
            let is_generic = instance
                .args
                .into_iter()
                .any(|arg| !matches!(arg.kind(), ty::GenericArgKind::Lifetime(_)));
            let can_be_inlined = matches!(
                ecx.tcx.sess.opts.unstable_opts.cross_crate_inline_threshold,
                InliningThreshold::Always
            ) || !matches!(
                ecx.tcx.codegen_instance_attrs(instance.def).inline,
                InlineAttr::Never
            );
            !is_generic && !can_be_inlined
        } else {
            // Non-functions are never unique.
            false
        };
        // Always use the same salt if the allocation is unique.
        if unique {
            CTFE_ALLOC_SALT
        } else {
            ecx.machine.rng.borrow_mut().random_range(0..ADDRS_PER_ANON_GLOBAL)
        }
    }

    fn cached_union_data_range<'e>(
        ecx: &'e mut InterpCx<'tcx, Self>,
        ty: Ty<'tcx>,
        compute_range: impl FnOnce() -> RangeSet,
    ) -> Cow<'e, RangeSet> {
        Cow::Borrowed(ecx.machine.union_data_ranges.entry(ty).or_insert_with(compute_range))
    }

    fn get_default_alloc_params(&self) -> <Self::Bytes as AllocBytes>::AllocParams {
        use crate::alloc::MiriAllocParams;

        match &self.allocator {
            Some(alloc) => MiriAllocParams::Isolated(alloc.clone()),
            None => MiriAllocParams::Global,
        }
    }

    fn enter_trace_span(span: impl FnOnce() -> tracing::Span) -> impl EnteredTraceSpan {
        #[cfg(feature = "tracing")]
        {
            span().entered()
        }
        #[cfg(not(feature = "tracing"))]
        #[expect(clippy::unused_unit)]
        {
            let _ = span; // so we avoid the "unused variable" warning
            ()
        }
    }
}

/// Trait for callbacks handling asynchronous machine operations.
pub trait MachineCallback<'tcx, T>: VisitProvenance {
    /// The function to be invoked when the callback is fired.
    fn call(
        self: Box<Self>,
        ecx: &mut InterpCx<'tcx, MiriMachine<'tcx>>,
        arg: T,
    ) -> InterpResult<'tcx>;
}

/// Type alias for boxed machine callbacks with generic argument type.
pub type DynMachineCallback<'tcx, T> = Box<dyn MachineCallback<'tcx, T> + 'tcx>;

/// Creates a `DynMachineCallback`:
///
/// ```rust
/// callback!(
///     @capture<'tcx> {
///         var1: Ty1,
///         var2: Ty2<'tcx>,
///     }
///     |this, arg: ArgTy| {
///         // Implement the callback here.
///         todo!()
///     }
/// )
/// ```
///
/// All the argument types must implement `VisitProvenance`.
#[macro_export]
macro_rules! callback {
    (@capture<$tcx:lifetime $(,)? $($lft:lifetime),*>
        { $($name:ident: $type:ty),* $(,)? }
     |$this:ident, $arg:ident: $arg_ty:ty| $body:expr $(,)?) => {{
        struct Callback<$tcx, $($lft),*> {
            $($name: $type,)*
            _phantom: std::marker::PhantomData<&$tcx ()>,
        }

        impl<$tcx, $($lft),*> VisitProvenance for Callback<$tcx, $($lft),*> {
            fn visit_provenance(&self, _visit: &mut VisitWith<'_>) {
                $(
                    VisitProvenance::visit_provenance(&self.$name, _visit);
                )*
            }
        }

        impl<$tcx, $($lft),*> MachineCallback<$tcx, $arg_ty> for Callback<$tcx, $($lft),*> {
            fn call(
                self: Box<Self>,
                $this: &mut MiriInterpCx<$tcx>,
                $arg: $arg_ty
            ) -> InterpResult<$tcx> {
                #[allow(unused_variables)]
                let Callback { $($name,)* _phantom } = *self;
                $body
            }
        }

        Box::new(Callback {
            $($name,)*
            _phantom: std::marker::PhantomData
        })
    }};
}