miri: optimize zeroed alloc

Co-authored-by: Ralf Jung <post@ralfj.de>

8 files changed, 59 insertions, 30 deletions

diff --git a/compiler/rustc_const_eval/src/const_eval/machine.rs b/compiler/rustc_const_eval/src/const_eval/machine.rs
index cfdfbdb7880..6a339d69542 100644
--- a/compiler/rustc_const_eval/src/const_eval/machine.rs
+++ b/compiler/rustc_const_eval/src/const_eval/machine.rs
@@ -21,9 +21,10 @@ use super::error::*;
 use crate::errors::{LongRunning, LongRunningWarn};
 use crate::fluent_generated as fluent;
 use crate::interpret::{
-    self, AllocId, AllocRange, ConstAllocation, CtfeProvenance, FnArg, Frame, GlobalAlloc, ImmTy,
-    InterpCx, InterpResult, MPlaceTy, OpTy, RangeSet, Scalar, compile_time_machine, interp_ok,
-    throw_exhaust, throw_inval, throw_ub, throw_ub_custom, throw_unsup, throw_unsup_format,
+    self, AllocId, AllocInit, AllocRange, ConstAllocation, CtfeProvenance, FnArg, Frame,
+    GlobalAlloc, ImmTy, InterpCx, InterpResult, MPlaceTy, OpTy, RangeSet, Scalar,
+    compile_time_machine, interp_ok, throw_exhaust, throw_inval, throw_ub, throw_ub_custom,
+    throw_unsup, throw_unsup_format,
 };
 
 /// When hitting this many interpreted terminators we emit a deny by default lint
@@ -420,6 +421,7 @@ impl<'tcx> interpret::Machine<'tcx> for CompileTimeMachine<'tcx> {
                     Size::from_bytes(size),
                     align,
                     interpret::MemoryKind::Machine(MemoryKind::Heap),
+                    AllocInit::Uninit,
                 )?;
                 ecx.write_pointer(ptr, dest)?;
             }
diff --git a/compiler/rustc_const_eval/src/interpret/memory.rs b/compiler/rustc_const_eval/src/interpret/memory.rs
index 2772c94d52b..1b8a222e9ea 100644
--- a/compiler/rustc_const_eval/src/interpret/memory.rs
+++ b/compiler/rustc_const_eval/src/interpret/memory.rs
@@ -20,10 +20,10 @@ use rustc_middle::ty::{self, Instance, Ty, TyCtxt};
 use tracing::{debug, instrument, trace};
 
 use super::{
-    AllocBytes, AllocId, AllocMap, AllocRange, Allocation, CheckAlignMsg, CheckInAllocMsg,
-    CtfeProvenance, GlobalAlloc, InterpCx, InterpResult, Machine, MayLeak, Misalignment, Pointer,
-    PointerArithmetic, Provenance, Scalar, alloc_range, err_ub, err_ub_custom, interp_ok, throw_ub,
-    throw_ub_custom, throw_unsup, throw_unsup_format,
+    AllocBytes, AllocId, AllocInit, AllocMap, AllocRange, Allocation, CheckAlignMsg,
+    CheckInAllocMsg, CtfeProvenance, GlobalAlloc, InterpCx, InterpResult, Machine, MayLeak,
+    Misalignment, Pointer, PointerArithmetic, Provenance, Scalar, alloc_range, err_ub,
+    err_ub_custom, interp_ok, throw_ub, throw_ub_custom, throw_unsup, throw_unsup_format,
 };
 use crate::fluent_generated as fluent;
 
@@ -230,11 +230,12 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         size: Size,
         align: Align,
         kind: MemoryKind<M::MemoryKind>,
+        init: AllocInit,
     ) -> InterpResult<'tcx, Pointer<M::Provenance>> {
         let alloc = if M::PANIC_ON_ALLOC_FAIL {
-            Allocation::uninit(size, align)
+            Allocation::new(size, align, init)
         } else {
-            Allocation::try_uninit(size, align)?
+            Allocation::try_new(size, align, init)?
         };
         self.insert_allocation(alloc, kind)
     }
@@ -270,6 +271,8 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         M::adjust_alloc_root_pointer(self, Pointer::from(id), Some(kind))
     }
 
+    /// If this grows the allocation, `init_growth` determines
+    /// whether the additional space will be initialized.
     pub fn reallocate_ptr(
         &mut self,
         ptr: Pointer<Option<M::Provenance>>,
@@ -277,6 +280,7 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         new_size: Size,
         new_align: Align,
         kind: MemoryKind<M::MemoryKind>,
+        init_growth: AllocInit,
     ) -> InterpResult<'tcx, Pointer<M::Provenance>> {
         let (alloc_id, offset, _prov) = self.ptr_get_alloc_id(ptr, 0)?;
         if offset.bytes() != 0 {
@@ -289,7 +293,10 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
 
         // For simplicities' sake, we implement reallocate as "alloc, copy, dealloc".
         // This happens so rarely, the perf advantage is outweighed by the maintenance cost.
-        let new_ptr = self.allocate_ptr(new_size, new_align, kind)?;
+        // If requested, we zero-init the entire allocation, to ensure that a growing
+        // allocation has its new bytes properly set. For the part that is copied,
+        // `mem_copy` below will de-initialize things as necessary.
+        let new_ptr = self.allocate_ptr(new_size, new_align, kind, init_growth)?;
         let old_size = match old_size_and_align {
             Some((size, _align)) => size,
             None => self.get_alloc_raw(alloc_id)?.size(),
diff --git a/compiler/rustc_const_eval/src/interpret/place.rs b/compiler/rustc_const_eval/src/interpret/place.rs
index c97922ac132..f5d3de7b1b2 100644
--- a/compiler/rustc_const_eval/src/interpret/place.rs
+++ b/compiler/rustc_const_eval/src/interpret/place.rs
@@ -12,9 +12,9 @@ use rustc_middle::{bug, mir, span_bug};
 use tracing::{instrument, trace};
 
 use super::{
-    AllocRef, AllocRefMut, CheckAlignMsg, CtfeProvenance, ImmTy, Immediate, InterpCx, InterpResult,
-    Machine, MemoryKind, Misalignment, OffsetMode, OpTy, Operand, Pointer, Projectable, Provenance,
-    Scalar, alloc_range, interp_ok, mir_assign_valid_types,
+    AllocInit, AllocRef, AllocRefMut, CheckAlignMsg, CtfeProvenance, ImmTy, Immediate, InterpCx,
+    InterpResult, Machine, MemoryKind, Misalignment, OffsetMode, OpTy, Operand, Pointer,
+    Projectable, Provenance, Scalar, alloc_range, interp_ok, mir_assign_valid_types,
 };
 
 #[derive(Copy, Clone, Hash, PartialEq, Eq, Debug)]
@@ -983,7 +983,7 @@ where
         let Some((size, align)) = self.size_and_align_of(&meta, &layout)? else {
             span_bug!(self.cur_span(), "cannot allocate space for `extern` type, size is not known")
         };
-        let ptr = self.allocate_ptr(size, align, kind)?;
+        let ptr = self.allocate_ptr(size, align, kind, AllocInit::Uninit)?;
         interp_ok(self.ptr_with_meta_to_mplace(ptr.into(), meta, layout, /*unaligned*/ false))
     }
 
diff --git a/compiler/rustc_const_eval/src/interpret/util.rs b/compiler/rustc_const_eval/src/interpret/util.rs
index ecb7c3fc93c..eb98e3b5380 100644
--- a/compiler/rustc_const_eval/src/interpret/util.rs
+++ b/compiler/rustc_const_eval/src/interpret/util.rs
@@ -2,7 +2,7 @@ use std::ops::ControlFlow;
 
 use rustc_hir::def_id::LocalDefId;
 use rustc_middle::mir;
-use rustc_middle::mir::interpret::{Allocation, InterpResult, Pointer};
+use rustc_middle::mir::interpret::{AllocInit, Allocation, InterpResult, Pointer};
 use rustc_middle::ty::layout::TyAndLayout;
 use rustc_middle::ty::{
     self, Ty, TyCtxt, TypeSuperVisitable, TypeVisitable, TypeVisitableExt, TypeVisitor,
@@ -76,7 +76,7 @@ pub(crate) fn create_static_alloc<'tcx>(
     static_def_id: LocalDefId,
     layout: TyAndLayout<'tcx>,
 ) -> InterpResult<'tcx, MPlaceTy<'tcx>> {
-    let alloc = Allocation::try_uninit(layout.size, layout.align.abi)?;
+    let alloc = Allocation::try_new(layout.size, layout.align.abi, AllocInit::Uninit)?;
     let alloc_id = ecx.tcx.reserve_and_set_static_alloc(static_def_id.into());
     assert_eq!(ecx.machine.static_root_ids, None);
     ecx.machine.static_root_ids = Some((alloc_id, static_def_id));
diff --git a/compiler/rustc_middle/src/mir/interpret/allocation.rs b/compiler/rustc_middle/src/mir/interpret/allocation.rs
index 1b07846e0cf..2e4d258ffff 100644
--- a/compiler/rustc_middle/src/mir/interpret/allocation.rs
+++ b/compiler/rustc_middle/src/mir/interpret/allocation.rs
@@ -270,6 +270,12 @@ impl AllocRange {
     }
 }
 
+/// Whether a new allocation should be initialized with zero-bytes.
+pub enum AllocInit {
+    Uninit,
+    Zero,
+}
+
 // The constructors are all without extra; the extra gets added by a machine hook later.
 impl<Prov: Provenance, Bytes: AllocBytes> Allocation<Prov, (), Bytes> {
     /// Creates an allocation initialized by the given bytes
@@ -294,7 +300,12 @@ impl<Prov: Provenance, Bytes: AllocBytes> Allocation<Prov, (), Bytes> {
         Allocation::from_bytes(slice, Align::ONE, Mutability::Not)
     }
 
-    fn uninit_inner<R>(size: Size, align: Align, fail: impl FnOnce() -> R) -> Result<Self, R> {
+    fn new_inner<R>(
+        size: Size,
+        align: Align,
+        init: AllocInit,
+        fail: impl FnOnce() -> R,
+    ) -> Result<Self, R> {
         // We raise an error if we cannot create the allocation on the host.
         // This results in an error that can happen non-deterministically, since the memory
         // available to the compiler can change between runs. Normally queries are always
@@ -306,7 +317,10 @@ impl<Prov: Provenance, Bytes: AllocBytes> Allocation<Prov, (), Bytes> {
         Ok(Allocation {
             bytes,
             provenance: ProvenanceMap::new(),
-            init_mask: InitMask::new(size, false),
+            init_mask: InitMask::new(size, match init {
+                AllocInit::Uninit => false,
+                AllocInit::Zero => true,
+            }),
             align,
             mutability: Mutability::Mut,
             extra: (),
@@ -315,8 +329,8 @@ impl<Prov: Provenance, Bytes: AllocBytes> Allocation<Prov, (), Bytes> {
 
     /// Try to create an Allocation of `size` bytes, failing if there is not enough memory
     /// available to the compiler to do so.
-    pub fn try_uninit<'tcx>(size: Size, align: Align) -> InterpResult<'tcx, Self> {
-        Self::uninit_inner(size, align, || {
+    pub fn try_new<'tcx>(size: Size, align: Align, init: AllocInit) -> InterpResult<'tcx, Self> {
+        Self::new_inner(size, align, init, || {
             ty::tls::with(|tcx| tcx.dcx().delayed_bug("exhausted memory during interpretation"));
             InterpErrorKind::ResourceExhaustion(ResourceExhaustionInfo::MemoryExhausted)
         })
@@ -328,8 +342,8 @@ impl<Prov: Provenance, Bytes: AllocBytes> Allocation<Prov, (), Bytes> {
     ///
     /// Example use case: To obtain an Allocation filled with specific data,
     /// first call this function and then call write_scalar to fill in the right data.
-    pub fn uninit(size: Size, align: Align) -> Self {
-        match Self::uninit_inner(size, align, || {
+    pub fn new(size: Size, align: Align, init: AllocInit) -> Self {
+        match Self::new_inner(size, align, init, || {
             panic!(
                 "interpreter ran out of memory: cannot create allocation of {} bytes",
                 size.bytes()
diff --git a/compiler/rustc_middle/src/mir/interpret/mod.rs b/compiler/rustc_middle/src/mir/interpret/mod.rs
index b88137544bc..f4f9f221a75 100644
--- a/compiler/rustc_middle/src/mir/interpret/mod.rs
+++ b/compiler/rustc_middle/src/mir/interpret/mod.rs
@@ -30,8 +30,8 @@ pub use {
 };
 
 pub use self::allocation::{
-    AllocBytes, AllocError, AllocRange, AllocResult, Allocation, ConstAllocation, InitChunk,
-    InitChunkIter, alloc_range,
+    AllocBytes, AllocError, AllocInit, AllocRange, AllocResult, Allocation, ConstAllocation,
+    InitChunk, InitChunkIter, alloc_range,
 };
 pub use self::error::{
     BadBytesAccess, CheckAlignMsg, CheckInAllocMsg, ErrorHandled, EvalStaticInitializerRawResult,
diff --git a/compiler/rustc_middle/src/ty/vtable.rs b/compiler/rustc_middle/src/ty/vtable.rs
index 23e2e8ad3d3..455bd16ff8c 100644
--- a/compiler/rustc_middle/src/ty/vtable.rs
+++ b/compiler/rustc_middle/src/ty/vtable.rs
@@ -4,7 +4,9 @@ use rustc_ast::Mutability;
 use rustc_macros::HashStable;
 use rustc_type_ir::elaborate;
 
-use crate::mir::interpret::{AllocId, Allocation, CTFE_ALLOC_SALT, Pointer, Scalar, alloc_range};
+use crate::mir::interpret::{
+    AllocId, AllocInit, Allocation, CTFE_ALLOC_SALT, Pointer, Scalar, alloc_range,
+};
 use crate::ty::{self, Instance, PolyTraitRef, Ty, TyCtxt};
 
 #[derive(Clone, Copy, PartialEq, HashStable)]
@@ -108,7 +110,7 @@ pub(super) fn vtable_allocation_provider<'tcx>(
     let ptr_align = tcx.data_layout.pointer_align.abi;
 
     let vtable_size = ptr_size * u64::try_from(vtable_entries.len()).unwrap();
-    let mut vtable = Allocation::uninit(vtable_size, ptr_align);
+    let mut vtable = Allocation::new(vtable_size, ptr_align, AllocInit::Uninit);
 
     // No need to do any alignment checks on the memory accesses below, because we know the
     // allocation is correctly aligned as we created it above. Also we're only offsetting by
diff --git a/compiler/rustc_smir/src/rustc_smir/alloc.rs b/compiler/rustc_smir/src/rustc_smir/alloc.rs
index 4e8db6096d4..52c5b425c14 100644
--- a/compiler/rustc_smir/src/rustc_smir/alloc.rs
+++ b/compiler/rustc_smir/src/rustc_smir/alloc.rs
@@ -1,6 +1,6 @@
 use rustc_abi::{Align, Size};
 use rustc_middle::mir::ConstValue;
-use rustc_middle::mir::interpret::{AllocRange, Pointer, alloc_range};
+use rustc_middle::mir::interpret::{AllocInit, AllocRange, Pointer, alloc_range};
 use stable_mir::Error;
 use stable_mir::mir::Mutability;
 use stable_mir::ty::{Allocation, ProvenanceMap};
@@ -44,7 +44,8 @@ pub(crate) fn try_new_allocation<'tcx>(
                 .layout_of(rustc_middle::ty::TypingEnv::fully_monomorphized().as_query_input(ty))
                 .map_err(|e| e.stable(tables))?
                 .align;
-            let mut allocation = rustc_middle::mir::interpret::Allocation::uninit(size, align.abi);
+            let mut allocation =
+                rustc_middle::mir::interpret::Allocation::new(size, align.abi, AllocInit::Uninit);
             allocation
                 .write_scalar(&tables.tcx, alloc_range(Size::ZERO, size), scalar)
                 .map_err(|e| e.stable(tables))?;
@@ -68,8 +69,11 @@ pub(crate) fn try_new_allocation<'tcx>(
                 .tcx
                 .layout_of(rustc_middle::ty::TypingEnv::fully_monomorphized().as_query_input(ty))
                 .map_err(|e| e.stable(tables))?;
-            let mut allocation =
-                rustc_middle::mir::interpret::Allocation::uninit(layout.size, layout.align.abi);
+            let mut allocation = rustc_middle::mir::interpret::Allocation::new(
+                layout.size,
+                layout.align.abi,
+                AllocInit::Uninit,
+            );
             allocation
                 .write_scalar(
                     &tables.tcx,