Auto merge of rust-lang#118640 - RalfJung:miri, r=RalfJung

Miri subtree update

needed to fix some errors in miri-test-libstd

Author: bors

src/tools/miri/rust-version

@@ -1 +1 @@
-225e36cff9809948d6567ab16f75d7b087ea83a7
+317d14a56cb8c748bf0e2f2afff89c2249ab4423

src/tools/miri/src/helpers.rs

@@ -1188,3 +1188,11 @@ pub(crate) fn simd_element_to_bool(elem: ImmTy<'_, Provenance>) -> InterpResult<
         _ => throw_ub_format!("each element of a SIMD mask must be all-0-bits or all-1-bits"),
     })
 }
+
+// This looks like something that would be nice to have in the standard library...
+pub(crate) fn round_to_next_multiple_of(x: u64, divisor: u64) -> u64 {
+    assert_ne!(divisor, 0);
+    // divisor is nonzero; multiplication cannot overflow since we just divided
+    #[allow(clippy::arithmetic_side_effects)]
+    return (x.checked_add(divisor - 1).unwrap() / divisor) * divisor;
+}

src/tools/miri/src/machine.rs

@@ -777,11 +777,6 @@ impl<'mir, 'tcx> MiriMachine<'mir, 'tcx> {
         drop(self.profiler.take());
     }
 
-    pub(crate) fn round_up_to_multiple_of_page_size(&self, length: u64) -> Option<u64> {
-        #[allow(clippy::arithmetic_side_effects)] // page size is nonzero
-        (length.checked_add(self.page_size - 1)? / self.page_size).checked_mul(self.page_size)
-    }
-
     pub(crate) fn page_align(&self) -> Align {
         Align::from_bytes(self.page_size).unwrap()
     }

src/tools/miri/src/shims/foreign_items.rs

@@ -558,16 +558,26 @@ trait EvalContextExtPriv<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
 
             // Promises that a pointer has a given symbolic alignment.
             "miri_promise_symbolic_alignment" => {
+                use rustc_target::abi::AlignFromBytesError;
+
                 let [ptr, align] = this.check_shim(abi, Abi::Rust, link_name, args)?;
                 let ptr = this.read_pointer(ptr)?;
                 let align = this.read_target_usize(align)?;
-                let Ok(align) = Align::from_bytes(align) else {
+                if !align.is_power_of_two() {
                     throw_unsup_format!(
-                        "`miri_promise_symbolic_alignment`: alignment must be a power of 2"
+                        "`miri_promise_symbolic_alignment`: alignment must be a power of 2, got {align}"
                     );
-                };
+                }
+                let align = Align::from_bytes(align).unwrap_or_else(|err| {
+                    match err {
+                        AlignFromBytesError::NotPowerOfTwo(_) => unreachable!(),
+                        // When the alignment is a power of 2 but too big, clamp it to MAX.
+                        AlignFromBytesError::TooLarge(_) => Align::MAX,
+                    }
+                });
                 let (_, addr) = ptr.into_parts(); // we know the offset is absolute
-                if addr.bytes() % align.bytes() != 0 {
+                // Cannot panic since `align` is a power of 2 and hence non-zero.
+                if addr.bytes().checked_rem(align.bytes()).unwrap() != 0 {
                     throw_unsup_format!(
                         "`miri_promise_symbolic_alignment`: pointer is not actually aligned"
                     );

src/tools/miri/src/shims/intrinsics/simd.rs

@@ -4,8 +4,10 @@ use rustc_middle::{mir, ty, ty::FloatTy};
 use rustc_span::{sym, Symbol};
 use rustc_target::abi::{Endian, HasDataLayout};
 
+use crate::helpers::{
+    bool_to_simd_element, check_arg_count, round_to_next_multiple_of, simd_element_to_bool,
+};
 use crate::*;
-use helpers::{bool_to_simd_element, check_arg_count, simd_element_to_bool};
 
 impl<'mir, 'tcx: 'mir> EvalContextExt<'mir, 'tcx> for crate::MiriInterpCx<'mir, 'tcx> {}
 pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
@@ -113,18 +115,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                             }
                         }
                         Op::Numeric(name) => {
-                            assert!(op.layout.ty.is_integral());
-                            let size = op.layout.size;
-                            let bits = op.to_scalar().to_bits(size).unwrap();
-                            let extra = 128u128.checked_sub(u128::from(size.bits())).unwrap();
-                            let bits_out = match name {
-                                sym::ctlz => u128::from(bits.leading_zeros()).checked_sub(extra).unwrap(),
-                                sym::cttz => u128::from((bits << extra).trailing_zeros()).checked_sub(extra).unwrap(),
-                                sym::bswap => (bits << extra).swap_bytes(),
-                                sym::bitreverse => (bits << extra).reverse_bits(),
-                                _ => unreachable!(),
-                            };
-                            Scalar::from_uint(bits_out, size)
+                            this.numeric_intrinsic(name, op.to_scalar(), op.layout)?
                         }
                     };
                     this.write_scalar(val, &dest)?;
@@ -405,37 +396,35 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                     this.write_immediate(*val, &dest)?;
                 }
             }
+            // Variant of `select` that takes a bitmask rather than a "vector of bool".
             "select_bitmask" => {
                 let [mask, yes, no] = check_arg_count(args)?;
                 let (yes, yes_len) = this.operand_to_simd(yes)?;
                 let (no, no_len) = this.operand_to_simd(no)?;
                 let (dest, dest_len) = this.place_to_simd(dest)?;
-                let bitmask_len = dest_len.max(8);
+                let bitmask_len = round_to_next_multiple_of(dest_len, 8);
 
+                // The mask must be an integer or an array.
+                assert!(
+                    mask.layout.ty.is_integral()
+                        || matches!(mask.layout.ty.kind(), ty::Array(elemty, _) if elemty == &this.tcx.types.u8)
+                );
                 assert!(bitmask_len <= 64);
                 assert_eq!(bitmask_len, mask.layout.size.bits());
                 assert_eq!(dest_len, yes_len);
                 assert_eq!(dest_len, no_len);
                 let dest_len = u32::try_from(dest_len).unwrap();
                 let bitmask_len = u32::try_from(bitmask_len).unwrap();
 
-                // The mask can be a single integer or an array.
-                let mask: u64 = match mask.layout.ty.kind() {
-                    ty::Int(..) | ty::Uint(..) =>
-                        this.read_scalar(mask)?.to_bits(mask.layout.size)?.try_into().unwrap(),
-                    ty::Array(elem, _) if matches!(elem.kind(), ty::Uint(ty::UintTy::U8)) => {
-                        let mask_ty = this.machine.layouts.uint(mask.layout.size).unwrap();
-                        let mask = mask.transmute(mask_ty, this)?;
-                        this.read_scalar(&mask)?.to_bits(mask_ty.size)?.try_into().unwrap()
-                    }
-                    _ => bug!("simd_select_bitmask: invalid mask type {}", mask.layout.ty),
-                };
+                // To read the mask, we transmute it to an integer.
+                // That does the right thing wrt endianess.
+                let mask_ty = this.machine.layouts.uint(mask.layout.size).unwrap();
+                let mask = mask.transmute(mask_ty, this)?;
+                let mask: u64 = this.read_scalar(&mask)?.to_bits(mask_ty.size)?.try_into().unwrap();
 
                 for i in 0..dest_len {
-                    let mask = mask
-                        & 1u64
-                            .checked_shl(simd_bitmask_index(i, dest_len, this.data_layout().endian))
-                            .unwrap();
+                    let bit_i = simd_bitmask_index(i, dest_len, this.data_layout().endian);
+                    let mask = mask & 1u64.checked_shl(bit_i).unwrap();
                     let yes = this.read_immediate(&this.project_index(&yes, i.into())?)?;
                     let no = this.read_immediate(&this.project_index(&no, i.into())?)?;
                     let dest = this.project_index(&dest, i.into())?;
@@ -445,6 +434,8 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                 }
                 for i in dest_len..bitmask_len {
                     // If the mask is "padded", ensure that padding is all-zero.
+                    // This deliberately does not use `simd_bitmask_index`; these bits are outside
+                    // the bitmask. It does not matter in which order we check them.
                     let mask = mask & 1u64.checked_shl(i).unwrap();
                     if mask != 0 {
                         throw_ub_format!(
@@ -453,6 +444,36 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                     }
                 }
             }
+            // Converts a "vector of bool" into a bitmask.
+            "bitmask" => {
+                let [op] = check_arg_count(args)?;
+                let (op, op_len) = this.operand_to_simd(op)?;
+                let bitmask_len = round_to_next_multiple_of(op_len, 8);
+
+                // Returns either an unsigned integer or array of `u8`.
+                assert!(
+                    dest.layout.ty.is_integral()
+                        || matches!(dest.layout.ty.kind(), ty::Array(elemty, _) if elemty == &this.tcx.types.u8)
+                );
+                assert!(bitmask_len <= 64);
+                assert_eq!(bitmask_len, dest.layout.size.bits());
+                let op_len = u32::try_from(op_len).unwrap();
+
+                let mut res = 0u64;
+                for i in 0..op_len {
+                    let op = this.read_immediate(&this.project_index(&op, i.into())?)?;
+                    if simd_element_to_bool(op)? {
+                        res |= 1u64
+                            .checked_shl(simd_bitmask_index(i, op_len, this.data_layout().endian))
+                            .unwrap();
+                    }
+                }
+                // We have to change the type of the place to be able to write `res` into it. This
+                // transmutes the integer to an array, which does the right thing wrt endianess.
+                let dest =
+                    dest.transmute(this.machine.layouts.uint(dest.layout.size).unwrap(), this)?;
+                this.write_int(res, &dest)?;
+            }
             "cast" | "as" | "cast_ptr" | "expose_addr" | "from_exposed_addr" => {
                 let [op] = check_arg_count(args)?;
                 let (op, op_len) = this.operand_to_simd(op)?;
@@ -635,34 +656,6 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                     }
                 }
             }
-            "bitmask" => {
-                let [op] = check_arg_count(args)?;
-                let (op, op_len) = this.operand_to_simd(op)?;
-                let bitmask_len = op_len.max(8);
-
-                // Returns either an unsigned integer or array of `u8`.
-                assert!(
-                    dest.layout.ty.is_integral()
-                        || matches!(dest.layout.ty.kind(), ty::Array(elemty, _) if elemty == &this.tcx.types.u8)
-                );
-                assert!(bitmask_len <= 64);
-                assert_eq!(bitmask_len, dest.layout.size.bits());
-                let op_len = u32::try_from(op_len).unwrap();
-
-                let mut res = 0u64;
-                for i in 0..op_len {
-                    let op = this.read_immediate(&this.project_index(&op, i.into())?)?;
-                    if simd_element_to_bool(op)? {
-                        res |= 1u64
-                            .checked_shl(simd_bitmask_index(i, op_len, this.data_layout().endian))
-                            .unwrap();
-                    }
-                }
-                // We have to force the place type to be an int so that we can write `res` into it.
-                let mut dest = this.force_allocation(dest)?;
-                dest.layout = this.machine.layouts.uint(dest.layout.size).unwrap();
-                this.write_int(res, &dest)?;
-            }
 
             name => throw_unsup_format!("unimplemented intrinsic: `simd_{name}`"),
         }

src/tools/miri/src/shims/unix/mem.rs

@@ -7,7 +7,7 @@
 //! equivalent. That is the only part we support: no MAP_FIXED or MAP_SHARED or anything
 //! else that goes beyond a basic allocation API.
 
-use crate::*;
+use crate::{helpers::round_to_next_multiple_of, *};
 use rustc_target::abi::Size;
 
 impl<'mir, 'tcx: 'mir> EvalContextExt<'mir, 'tcx> for crate::MiriInterpCx<'mir, 'tcx> {}
@@ -89,7 +89,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
         }
 
         let align = this.machine.page_align();
-        let map_length = this.machine.round_up_to_multiple_of_page_size(length).unwrap_or(u64::MAX);
+        let map_length = round_to_next_multiple_of(length, this.machine.page_size);
 
         let ptr =
             this.allocate_ptr(Size::from_bytes(map_length), align, MiriMemoryKind::Mmap.into())?;
@@ -123,7 +123,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
             return Ok(Scalar::from_i32(-1));
         }
 
-        let length = this.machine.round_up_to_multiple_of_page_size(length).unwrap_or(u64::MAX);
+        let length = round_to_next_multiple_of(length, this.machine.page_size);
 
         let ptr = Machine::ptr_from_addr_cast(this, addr)?;
 

src/tools/miri/tests/fail/unaligned_pointers/promise_alignment.rs

@@ -17,11 +17,7 @@ fn main() {
     let _val = unsafe { buffer.read() };
 
     // Let's find a place to promise alignment 8.
-    let align8 = if buffer.addr() % 8 == 0 {
-        buffer
-    } else {
-        buffer.wrapping_add(1)
-    };
+    let align8 = if buffer.addr() % 8 == 0 { buffer } else { buffer.wrapping_add(1) };
     assert!(align8.addr() % 8 == 0);
     unsafe { utils::miri_promise_symbolic_alignment(align8.cast(), 8) };
     // Promising the alignment down to 1 *again* still must not hurt.
@@ -41,11 +37,7 @@ fn main() {
         #[derive(Copy, Clone)]
         struct Align16(u128);
 
-        let align16 = if align8.addr() % 16 == 0 {
-            align8
-        } else {
-            align8.wrapping_add(2)
-        };
+        let align16 = if align8.addr() % 16 == 0 { align8 } else { align8.wrapping_add(2) };
         assert!(align16.addr() % 16 == 0);
 
         let _val = unsafe { align8.cast::<Align16>().read() };

src/tools/miri/tests/fail/unaligned_pointers/promise_alignment_zero.rs

@@ -0,0 +1,8 @@
+#[path = "../../utils/mod.rs"]
+mod utils;
+
+fn main() {
+    let buffer = [0u32; 128];
+    unsafe { utils::miri_promise_symbolic_alignment(buffer.as_ptr().cast(), 0) };
+    //~^ERROR: alignment must be a power of 2
+}

src/tools/miri/tests/fail/unaligned_pointers/promise_alignment_zero.stderr

@@ -0,0 +1,14 @@
+error: unsupported operation: `miri_promise_symbolic_alignment`: alignment must be a power of 2, got 0
+  --> $DIR/promise_alignment_zero.rs:LL:CC
+   |
+LL |     unsafe { utils::miri_promise_symbolic_alignment(buffer.as_ptr().cast(), 0) };
+   |              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ `miri_promise_symbolic_alignment`: alignment must be a power of 2, got 0
+   |
+   = help: this is likely not a bug in the program; it indicates that the program performed an operation that the interpreter does not support
+   = note: BACKTRACE:
+   = note: inside `main` at $DIR/promise_alignment_zero.rs:LL:CC
+
+note: some details are omitted, run with `MIRIFLAGS=-Zmiri-backtrace=full` for a verbose backtrace
+
+error: aborting due to 1 previous error
+

src/tools/miri/tests/pass/align_offset_symbolic.rs

@@ -90,9 +90,21 @@ fn test_cstr() {
     std::ffi::CStr::from_bytes_with_nul(b"this is a test that is longer than 16 bytes\0").unwrap();
 }
 
+fn huge_align() {
+    #[cfg(target_pointer_width = "64")]
+    const SIZE: usize = 1 << 47;
+    #[cfg(target_pointer_width = "32")]
+    const SIZE: usize = 1 << 30;
+    #[cfg(target_pointer_width = "16")]
+    const SIZE: usize = 1 << 13;
+    struct HugeSize([u8; SIZE - 1]);
+    let _ = std::ptr::invalid::<HugeSize>(SIZE).align_offset(SIZE);
+}
+
 fn main() {
     test_align_to();
     test_from_utf8();
     test_u64_array();
     test_cstr();
+    huge_align();
 }