Auto merge of #133950 - matthiaskrgr:rollup-b7g5p73, r=matthiaskrgr

Rollup of 5 pull requests

Successful merges:

 - #130777 (rust_for_linux: -Zreg-struct-return commandline flag for X86 (#116973))
 - #133211 (Extend Miri to correctly pass mutable pointers through FFI)
 - #133790 (Improve documentation for Vec::extend_from_within)
 - #133930 (rustbook: update to use new mdbook-trpl package from The Book)
 - #133931 (Only allow PassMode::Direct for aggregates on wasm when using the C ABI)

r? `@ghost`
`@rustbot` modify labels: rollup

Author: bors

compiler/rustc_codegen_gcc/src/context.rs

@@ -544,7 +544,10 @@ impl<'gcc, 'tcx> HasWasmCAbiOpt for CodegenCx<'gcc, 'tcx> {
 
 impl<'gcc, 'tcx> HasX86AbiOpt for CodegenCx<'gcc, 'tcx> {
     fn x86_abi_opt(&self) -> X86Abi {
-        X86Abi { regparm: self.tcx.sess.opts.unstable_opts.regparm }
+        X86Abi {
+            regparm: self.tcx.sess.opts.unstable_opts.regparm,
+            reg_struct_return: self.tcx.sess.opts.unstable_opts.reg_struct_return,
+        }
     }
 }
 

compiler/rustc_const_eval/src/const_eval/dummy_machine.rs

@@ -168,9 +168,9 @@ impl<'tcx> interpret::Machine<'tcx> for DummyMachine {
         })
     }
 
-    fn expose_ptr(
-        _ecx: &mut InterpCx<'tcx, Self>,
-        _ptr: interpret::Pointer<Self::Provenance>,
+    fn expose_provenance(
+        _ecx: &InterpCx<'tcx, Self>,
+        _provenance: Self::Provenance,
     ) -> interpret::InterpResult<'tcx> {
         unimplemented!()
     }

compiler/rustc_const_eval/src/const_eval/machine.rs

@@ -21,9 +21,8 @@ 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, Pointer, RangeSet, Scalar, compile_time_machine,
-    interp_ok, throw_exhaust, throw_inval, throw_ub, throw_ub_custom, throw_unsup,
-    throw_unsup_format,
+    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
@@ -586,7 +585,10 @@ impl<'tcx> interpret::Machine<'tcx> for CompileTimeMachine<'tcx> {
     }
 
     #[inline(always)]
-    fn expose_ptr(_ecx: &mut InterpCx<'tcx, Self>, _ptr: Pointer) -> InterpResult<'tcx> {
+    fn expose_provenance(
+        _ecx: &InterpCx<'tcx, Self>,
+        _provenance: Self::Provenance,
+    ) -> InterpResult<'tcx> {
         // This is only reachable with -Zunleash-the-miri-inside-of-you.
         throw_unsup_format!("exposing pointers is not possible at compile-time")
     }

compiler/rustc_const_eval/src/interpret/cast.rs

@@ -238,7 +238,7 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         let scalar = src.to_scalar();
         let ptr = scalar.to_pointer(self)?;
         match ptr.into_pointer_or_addr() {
-            Ok(ptr) => M::expose_ptr(self, ptr)?,
+            Ok(ptr) => M::expose_provenance(self, ptr.provenance)?,
             Err(_) => {} // Do nothing, exposing an invalid pointer (`None` provenance) is a NOP.
         };
         interp_ok(ImmTy::from_scalar(

compiler/rustc_const_eval/src/interpret/machine.rs

@@ -327,11 +327,11 @@ pub trait Machine<'tcx>: Sized {
         addr: u64,
     ) -> InterpResult<'tcx, Pointer<Option<Self::Provenance>>>;
 
-    /// Marks a pointer as exposed, allowing it's provenance
+    /// Marks a pointer as exposed, allowing its provenance
     /// to be recovered. "Pointer-to-int cast"
-    fn expose_ptr(
-        ecx: &mut InterpCx<'tcx, Self>,
-        ptr: Pointer<Self::Provenance>,
+    fn expose_provenance(
+        ecx: &InterpCx<'tcx, Self>,
+        provenance: Self::Provenance,
     ) -> InterpResult<'tcx>;
 
     /// Convert a pointer with provenance into an allocation-offset pair and extra provenance info.

compiler/rustc_const_eval/src/interpret/memory.rs

@@ -944,6 +944,52 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         interp_ok(())
     }
 
+    /// Handle the effect an FFI call might have on the state of allocations.
+    /// This overapproximates the modifications which external code might make to memory:
+    /// We set all reachable allocations as initialized, mark all provenances as exposed
+    /// and overwrite them with `Provenance::WILDCARD`.
+    pub fn prepare_for_native_call(
+        &mut self,
+        id: AllocId,
+        initial_prov: M::Provenance,
+    ) -> InterpResult<'tcx> {
+        // Expose provenance of the root allocation.
+        M::expose_provenance(self, initial_prov)?;
+
+        let mut done = FxHashSet::default();
+        let mut todo = vec![id];
+        while let Some(id) = todo.pop() {
+            if !done.insert(id) {
+                // We already saw this allocation before, don't process it again.
+                continue;
+            }
+            let info = self.get_alloc_info(id);
+
+            // If there is no data behind this pointer, skip this.
+            if !matches!(info.kind, AllocKind::LiveData) {
+                continue;
+            }
+
+            // Expose all provenances in this allocation, and add them to `todo`.
+            let alloc = self.get_alloc_raw(id)?;
+            for prov in alloc.provenance().provenances() {
+                M::expose_provenance(self, prov)?;
+                if let Some(id) = prov.get_alloc_id() {
+                    todo.push(id);
+                }
+            }
+
+            // Prepare for possible write from native code if mutable.
+            if info.mutbl.is_mut() {
+                self.get_alloc_raw_mut(id)?
+                    .0
+                    .prepare_for_native_write()
+                    .map_err(|e| e.to_interp_error(id))?;
+            }
+        }
+        interp_ok(())
+    }
+
     /// Create a lazy debug printer that prints the given allocation and all allocations it points
     /// to, recursively.
     #[must_use]

compiler/rustc_interface/src/tests.rs

@@ -832,6 +832,7 @@ fn test_unstable_options_tracking_hash() {
     tracked!(precise_enum_drop_elaboration, false);
     tracked!(profile_sample_use, Some(PathBuf::from("abc")));
     tracked!(profiler_runtime, "abc".to_string());
+    tracked!(reg_struct_return, true);
     tracked!(regparm, Some(3));
     tracked!(relax_elf_relocations, Some(true));
     tracked!(remap_cwd_prefix, Some(PathBuf::from("abc")));

compiler/rustc_middle/src/mir/interpret/allocation.rs

@@ -643,6 +643,28 @@ impl<Prov: Provenance, Extra, Bytes: AllocBytes> Allocation<Prov, Extra, Bytes>
         Ok(())
     }
 
+    /// Initialize all previously uninitialized bytes in the entire allocation, and set
+    /// provenance of everything to `Wildcard`. Before calling this, make sure all
+    /// provenance in this allocation is exposed!
+    pub fn prepare_for_native_write(&mut self) -> AllocResult {
+        let full_range = AllocRange { start: Size::ZERO, size: Size::from_bytes(self.len()) };
+        // Overwrite uninitialized bytes with 0, to ensure we don't leak whatever their value happens to be.
+        for chunk in self.init_mask.range_as_init_chunks(full_range) {
+            if !chunk.is_init() {
+                let uninit_bytes = &mut self.bytes
+                    [chunk.range().start.bytes_usize()..chunk.range().end.bytes_usize()];
+                uninit_bytes.fill(0);
+            }
+        }
+        // Mark everything as initialized now.
+        self.mark_init(full_range, true);
+
+        // Set provenance of all bytes to wildcard.
+        self.provenance.write_wildcards(self.len());
+
+        Ok(())
+    }
+
     /// Remove all provenance in the given memory range.
     pub fn clear_provenance(&mut self, cx: &impl HasDataLayout, range: AllocRange) -> AllocResult {
         self.provenance.clear(range, cx)?;

compiler/rustc_middle/src/mir/interpret/allocation/provenance_map.rs

@@ -195,6 +195,25 @@ impl<Prov: Provenance> ProvenanceMap<Prov> {
 
         Ok(())
     }
+
+    /// Overwrites all provenance in the allocation with wildcard provenance.
+    ///
+    /// Provided for usage in Miri and panics otherwise.
+    pub fn write_wildcards(&mut self, alloc_size: usize) {
+        assert!(
+            Prov::OFFSET_IS_ADDR,
+            "writing wildcard provenance is not supported when `OFFSET_IS_ADDR` is false"
+        );
+        let wildcard = Prov::WILDCARD.unwrap();
+
+        // Remove all pointer provenances, then write wildcards into the whole byte range.
+        self.ptrs.clear();
+        let last = Size::from_bytes(alloc_size);
+        let bytes = self.bytes.get_or_insert_with(Box::default);
+        for offset in Size::ZERO..last {
+            bytes.insert(offset, wildcard);
+        }
+    }
 }
 
 /// A partial, owned list of provenance to transfer into another allocation.

compiler/rustc_middle/src/mir/interpret/pointer.rs

@@ -66,6 +66,9 @@ pub trait Provenance: Copy + fmt::Debug + 'static {
     ///   pointer, and implement ptr-to-int transmutation by stripping provenance.
     const OFFSET_IS_ADDR: bool;
 
+    /// If wildcard provenance is implemented, contains the unique, general wildcard provenance variant.
+    const WILDCARD: Option<Self>;
+
     /// Determines how a pointer should be printed.
     fn fmt(ptr: &Pointer<Self>, f: &mut fmt::Formatter<'_>) -> fmt::Result;
 
@@ -168,6 +171,9 @@ impl Provenance for CtfeProvenance {
     // so ptr-to-int casts are not possible (since we do not know the global physical offset).
     const OFFSET_IS_ADDR: bool = false;
 
+    // `CtfeProvenance` does not implement wildcard provenance.
+    const WILDCARD: Option<Self> = None;
+
     fn fmt(ptr: &Pointer<Self>, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         // Print AllocId.
         fmt::Debug::fmt(&ptr.provenance.alloc_id(), f)?; // propagates `alternate` flag
@@ -197,6 +203,9 @@ impl Provenance for AllocId {
     // so ptr-to-int casts are not possible (since we do not know the global physical offset).
     const OFFSET_IS_ADDR: bool = false;
 
+    // `AllocId` does not implement wildcard provenance.
+    const WILDCARD: Option<Self> = None;
+
     fn fmt(ptr: &Pointer<Self>, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         // Forward `alternate` flag to `alloc_id` printing.
         if f.alternate() {

compiler/rustc_middle/src/ty/layout.rs

@@ -552,7 +552,10 @@ impl<'tcx> HasWasmCAbiOpt for TyCtxt<'tcx> {
 
 impl<'tcx> HasX86AbiOpt for TyCtxt<'tcx> {
     fn x86_abi_opt(&self) -> X86Abi {
-        X86Abi { regparm: self.sess.opts.unstable_opts.regparm }
+        X86Abi {
+            regparm: self.sess.opts.unstable_opts.regparm,
+            reg_struct_return: self.sess.opts.unstable_opts.reg_struct_return,
+        }
     }
 }
 

compiler/rustc_session/messages.ftl

@@ -135,5 +135,6 @@ session_unsupported_crate_type_for_target =
 
 session_unsupported_dwarf_version = requested DWARF version {$dwarf_version} is greater than 5
 
+session_unsupported_reg_struct_return_arch = `-Zreg-struct-return` is only supported on x86
 session_unsupported_regparm = `-Zregparm={$regparm}` is unsupported (valid values 0-3)
 session_unsupported_regparm_arch = `-Zregparm=N` is only supported on x86

compiler/rustc_session/src/errors.rs

@@ -489,6 +489,10 @@ pub(crate) struct UnsupportedRegparm {
 #[diag(session_unsupported_regparm_arch)]
 pub(crate) struct UnsupportedRegparmArch;
 
+#[derive(Diagnostic)]
+#[diag(session_unsupported_reg_struct_return_arch)]
+pub(crate) struct UnsupportedRegStructReturnArch;
+
 #[derive(Diagnostic)]
 #[diag(session_failed_to_create_profiler)]
 pub(crate) struct FailedToCreateProfiler {

compiler/rustc_session/src/options.rs

@@ -1988,6 +1988,9 @@ options! {
         "enable queries of the dependency graph for regression testing (default: no)"),
     randomize_layout: bool = (false, parse_bool, [TRACKED],
         "randomize the layout of types (default: no)"),
+    reg_struct_return: bool = (false, parse_bool, [TRACKED],
+        "On x86-32 targets, it overrides the default ABI to return small structs in registers.
+        It is UNSOUND to link together crates that use different values for this flag!"),
     regparm: Option<u32> = (None, parse_opt_number, [TRACKED],
         "On x86-32 targets, setting this to N causes the compiler to pass N arguments \
         in registers EAX, EDX, and ECX instead of on the stack for\

compiler/rustc_session/src/session.rs

@@ -1305,6 +1305,11 @@ fn validate_commandline_args_with_session_available(sess: &Session) {
             sess.dcx().emit_err(errors::UnsupportedRegparmArch);
         }
     }
+    if sess.opts.unstable_opts.reg_struct_return {
+        if sess.target.arch != "x86" {
+            sess.dcx().emit_err(errors::UnsupportedRegStructReturnArch);
+        }
+    }
 
     // The code model check applies to `thunk` and `thunk-extern`, but not `thunk-inline`, so it is
     // kept as a `match` to force a change if new ones are added, even if we currently only support

compiler/rustc_target/src/callconv/mod.rs

@@ -661,7 +661,9 @@ impl<'a, Ty> FnAbi<'a, Ty> {
                     }
                     _ => (x86::Flavor::General, None),
                 };
-                x86::compute_abi_info(cx, self, x86::X86Options { flavor, regparm });
+                let reg_struct_return = cx.x86_abi_opt().reg_struct_return;
+                let opts = x86::X86Options { flavor, regparm, reg_struct_return };
+                x86::compute_abi_info(cx, self, opts);
             }
             "x86_64" => match abi {
                 spec::abi::Abi::SysV64 { .. } => x86_64::compute_abi_info(cx, self),

compiler/rustc_target/src/callconv/x86.rs

@@ -14,6 +14,7 @@ pub(crate) enum Flavor {
 pub(crate) struct X86Options {
     pub flavor: Flavor,
     pub regparm: Option<u32>,
+    pub reg_struct_return: bool,
 }
 
 pub(crate) fn compute_abi_info<'a, Ty, C>(cx: &C, fn_abi: &mut FnAbi<'a, Ty>, opts: X86Options)
@@ -31,7 +32,7 @@ where
             // https://www.angelcode.com/dev/callconv/callconv.html
             // Clang's ABI handling is in lib/CodeGen/TargetInfo.cpp
             let t = cx.target_spec();
-            if t.abi_return_struct_as_int {
+            if t.abi_return_struct_as_int || opts.reg_struct_return {
                 // According to Clang, everyone but MSVC returns single-element
                 // float aggregates directly in a floating-point register.
                 if !t.is_like_msvc && fn_abi.ret.layout.is_single_fp_element(cx) {

compiler/rustc_target/src/spec/mod.rs

@@ -2117,6 +2117,8 @@ pub struct X86Abi {
     /// On x86-32 targets, the regparm N causes the compiler to pass arguments
     /// in registers EAX, EDX, and ECX instead of on the stack.
     pub regparm: Option<u32>,
+    /// Override the default ABI to return small structs in registers
+    pub reg_struct_return: bool,
 }
 
 pub trait HasX86AbiOpt {

compiler/rustc_ty_utils/src/abi.rs

@@ -473,20 +473,30 @@ fn fn_abi_sanity_check<'tcx>(
                         // This really shouldn't happen even for sized aggregates, since
                         // `immediate_llvm_type` will use `layout.fields` to turn this Rust type into an
                         // LLVM type. This means all sorts of Rust type details leak into the ABI.
-                        // However wasm sadly *does* currently use this mode so we have to allow it --
-                        // but we absolutely shouldn't let any more targets do that.
-                        // (Also see <https://github.com/rust-lang/rust/issues/115666>.)
+                        // However wasm sadly *does* currently use this mode for it's "C" ABI so we
+                        // have to allow it -- but we absolutely shouldn't let any more targets do
+                        // that. (Also see <https://github.com/rust-lang/rust/issues/115666>.)
                         //
                         // The unstable abi `PtxKernel` also uses Direct for now.
                         // It needs to switch to something else before stabilization can happen.
                         // (See issue: https://github.com/rust-lang/rust/issues/117271)
-                        assert!(
-                            matches!(&*tcx.sess.target.arch, "wasm32" | "wasm64")
-                                || matches!(spec_abi, ExternAbi::PtxKernel | ExternAbi::Unadjusted),
-                            "`PassMode::Direct` for aggregates only allowed for \"unadjusted\" and \"ptx-kernel\" functions and on wasm\n\
-                          Problematic type: {:#?}",
-                            arg.layout,
-                        );
+                        //
+                        // And finally the unadjusted ABI is ill specified and uses Direct for all
+                        // args, but unfortunately we need it for calling certain LLVM intrinsics.
+
+                        match spec_abi {
+                            ExternAbi::Unadjusted => {}
+                            ExternAbi::PtxKernel => {}
+                            ExternAbi::C { unwind: _ }
+                                if matches!(&*tcx.sess.target.arch, "wasm32" | "wasm64") => {}
+                            _ => {
+                                panic!(
+                                    "`PassMode::Direct` for aggregates only allowed for \"unadjusted\" and \"ptx-kernel\" functions and on wasm\n\
+                                      Problematic type: {:#?}",
+                                    arg.layout,
+                                );
+                            }
+                        }
                     }
                 }
             }

library/alloc/src/vec/mod.rs

@@ -3025,26 +3025,29 @@ impl<T: Clone, A: Allocator> Vec<T, A> {
         self.spec_extend(other.iter())
     }
 
-    /// Copies elements from `src` range to the end of the vector.
+    /// Given a range `src`, clones a slice of elements in that range and appends it to the end.
+    ///
+    /// `src` must be a range that can form a valid subslice of the `Vec`.
     ///
     /// # Panics
     ///
-    /// Panics if the starting point is greater than the end point or if
-    /// the end point is greater than the length of the vector.
+    /// Panics if starting index is greater than the end index
+    /// or if the index is greater than the length of the vector.
     ///
     /// # Examples
     ///
     /// ```
-    /// let mut vec = vec![0, 1, 2, 3, 4];
-    ///
-    /// vec.extend_from_within(2..);
-    /// assert_eq!(vec, [0, 1, 2, 3, 4, 2, 3, 4]);
+    /// let mut characters = vec!['a', 'b', 'c', 'd', 'e'];
+    /// characters.extend_from_within(2..);
+    /// assert_eq!(characters, ['a', 'b', 'c', 'd', 'e', 'c', 'd', 'e']);
     ///
-    /// vec.extend_from_within(..2);
-    /// assert_eq!(vec, [0, 1, 2, 3, 4, 2, 3, 4, 0, 1]);
+    /// let mut numbers = vec![0, 1, 2, 3, 4];
+    /// numbers.extend_from_within(..2);
+    /// assert_eq!(numbers, [0, 1, 2, 3, 4, 0, 1]);
     ///
-    /// vec.extend_from_within(4..8);
-    /// assert_eq!(vec, [0, 1, 2, 3, 4, 2, 3, 4, 0, 1, 4, 2, 3, 4]);
+    /// let mut strings = vec![String::from("hello"), String::from("world"), String::from("!")];
+    /// strings.extend_from_within(1..=2);
+    /// assert_eq!(strings, ["hello", "world", "!", "world", "!"]);
     /// ```
     #[cfg(not(no_global_oom_handling))]
     #[stable(feature = "vec_extend_from_within", since = "1.53.0")]