Auto merge of #118538 - RalfJung:size-of-val-comments, r=WaffleLapkin

fix dynamic size/align computation logic for packed types with dyn trait tail

This logic was never updated to support `packed(N)` where `N > 1`, and it turns out to be wrong for that case.

Fixes https://github.com/rust-lang/rust/issues/80925

`@bjorn3` I have not looked at cranelift; I assume it basically copied the size-of-val logic and hence could use much the same patch.
This commit is contained in:
bors 2023-12-14 07:19:07 +00:00
commit 1a8afa0e74
4 changed files with 98 additions and 47 deletions

View File

@ -84,10 +84,13 @@ pub fn size_and_align_of_dst<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
debug!("DST {} layout: {:?}", t, layout);
let i = layout.fields.count() - 1;
let sized_size = layout.fields.offset(i).bytes();
let unsized_offset_unadjusted = layout.fields.offset(i).bytes();
let sized_align = layout.align.abi.bytes();
debug!("DST {} statically sized prefix size: {} align: {}", t, sized_size, sized_align);
let sized_size = bx.const_usize(sized_size);
debug!(
"DST {} offset of dyn field: {}, statically sized align: {}",
t, unsized_offset_unadjusted, sized_align
);
let unsized_offset_unadjusted = bx.const_usize(unsized_offset_unadjusted);
let sized_align = bx.const_usize(sized_align);
// Recurse to get the size of the dynamically sized field (must be
@ -95,26 +98,26 @@ pub fn size_and_align_of_dst<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
let field_ty = layout.field(bx, i).ty;
let (unsized_size, mut unsized_align) = size_and_align_of_dst(bx, field_ty, info);
// FIXME (#26403, #27023): We should be adding padding
// to `sized_size` (to accommodate the `unsized_align`
// required of the unsized field that follows) before
// summing it with `sized_size`. (Note that since #26403
// is unfixed, we do not yet add the necessary padding
// here. But this is where the add would go.)
// # First compute the dynamic alignment
// Return the sum of sizes and max of aligns.
let size = bx.add(sized_size, unsized_size);
// Packed types ignore the alignment of their fields.
if let ty::Adt(def, _) = t.kind() {
if def.repr().packed() {
unsized_align = sized_align;
// For packed types, we need to cap the alignment.
if let ty::Adt(def, _) = t.kind()
&& let Some(packed) = def.repr().pack
{
if packed.bytes() == 1 {
// We know this will be capped to 1.
unsized_align = bx.const_usize(1);
} else {
// We have to dynamically compute `min(unsized_align, packed)`.
let packed = bx.const_usize(packed.bytes());
let cmp = bx.icmp(IntPredicate::IntULT, unsized_align, packed);
unsized_align = bx.select(cmp, unsized_align, packed);
}
}
// Choose max of two known alignments (combined value must
// be aligned according to more restrictive of the two).
let align = match (
let full_align = match (
bx.const_to_opt_u128(sized_align, false),
bx.const_to_opt_u128(unsized_align, false),
) {
@ -129,6 +132,19 @@ pub fn size_and_align_of_dst<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
}
};
// # Then compute the dynamic size
// The full formula for the size would be:
// let unsized_offset_adjusted = unsized_offset_unadjusted.align_to(unsized_align);
// let full_size = (unsized_offset_adjusted + unsized_size).align_to(full_align);
// However, `unsized_size` is a multiple of `unsized_align`.
// Therefore, we can equivalently do the `align_to(unsized_align)` *after* adding `unsized_size`:
// let full_size = (unsized_offset_unadjusted + unsized_size).align_to(unsized_align).align_to(full_align);
// Furthermore, `align >= unsized_align`, and therefore we only need to do:
// let full_size = (unsized_offset_unadjusted + unsized_size).align_to(full_align);
let full_size = bx.add(unsized_offset_unadjusted, unsized_size);
// Issue #27023: must add any necessary padding to `size`
// (to make it a multiple of `align`) before returning it.
//
@ -140,12 +156,12 @@ pub fn size_and_align_of_dst<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
//
// `(size + (align-1)) & -align`
let one = bx.const_usize(1);
let addend = bx.sub(align, one);
let add = bx.add(size, addend);
let neg = bx.neg(align);
let size = bx.and(add, neg);
let addend = bx.sub(full_align, one);
let add = bx.add(full_size, addend);
let neg = bx.neg(full_align);
let full_size = bx.and(add, neg);
(size, align)
(full_size, full_align)
}
_ => bug!("size_and_align_of_dst: {t} not supported"),
}

View File

@ -686,14 +686,8 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
assert!(layout.fields.count() > 0);
trace!("DST layout: {:?}", layout);
let sized_size = layout.fields.offset(layout.fields.count() - 1);
let unsized_offset_unadjusted = layout.fields.offset(layout.fields.count() - 1);
let sized_align = layout.align.abi;
trace!(
"DST {} statically sized prefix size: {:?} align: {:?}",
layout.ty,
sized_size,
sized_align
);
// Recurse to get the size of the dynamically sized field (must be
// the last field). Can't have foreign types here, how would we
@ -707,36 +701,35 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
return Ok(None);
};
// FIXME (#26403, #27023): We should be adding padding
// to `sized_size` (to accommodate the `unsized_align`
// required of the unsized field that follows) before
// summing it with `sized_size`. (Note that since #26403
// is unfixed, we do not yet add the necessary padding
// here. But this is where the add would go.)
// # First compute the dynamic alignment
// Return the sum of sizes and max of aligns.
let size = sized_size + unsized_size; // `Size` addition
// Packed types ignore the alignment of their fields.
// Packed type alignment needs to be capped.
if let ty::Adt(def, _) = layout.ty.kind() {
if def.repr().packed() {
unsized_align = sized_align;
if let Some(packed) = def.repr().pack {
unsized_align = unsized_align.min(packed);
}
}
// Choose max of two known alignments (combined value must
// be aligned according to more restrictive of the two).
let align = sized_align.max(unsized_align);
let full_align = sized_align.max(unsized_align);
// Issue #27023: must add any necessary padding to `size`
// (to make it a multiple of `align`) before returning it.
let size = size.align_to(align);
// # Then compute the dynamic size
let unsized_offset_adjusted = unsized_offset_unadjusted.align_to(unsized_align);
let full_size = (unsized_offset_adjusted + unsized_size).align_to(full_align);
// Just for our sanitiy's sake, assert that this is equal to what codegen would compute.
assert_eq!(
full_size,
(unsized_offset_unadjusted + unsized_size).align_to(full_align)
);
// Check if this brought us over the size limit.
if size > self.max_size_of_val() {
if full_size > self.max_size_of_val() {
throw_ub!(InvalidMeta(InvalidMetaKind::TooBig));
}
Ok(Some((size, align)))
Ok(Some((full_size, full_align)))
}
ty::Dynamic(_, _, ty::Dyn) => {
let vtable = metadata.unwrap_meta().to_pointer(self)?;

View File

@ -0,0 +1,21 @@
// run-pass
use std::ptr::addr_of;
// When the unsized tail is a `dyn Trait`, its alignments is only dynamically known. This means the
// packed(2) needs to be applied at runtime: the actual alignment of the field is `min(2,
// usual_alignment)`. Here we check that we do this right by comparing size, alignment, and field
// offset before and after unsizing.
fn main() {
#[repr(C, packed(2))]
struct Packed<T: ?Sized>(u8, core::mem::ManuallyDrop<T>);
let p = Packed(0, core::mem::ManuallyDrop::new(1));
let p: &Packed<usize> = &p;
let sized = (core::mem::size_of_val(p), core::mem::align_of_val(p));
let sized_offset = unsafe { addr_of!(p.1).cast::<u8>().offset_from(addr_of!(p.0)) };
let p: &Packed<dyn Send> = p;
let un_sized = (core::mem::size_of_val(p), core::mem::align_of_val(p));
let un_sized_offset = unsafe { addr_of!(p.1).cast::<u8>().offset_from(addr_of!(p.0)) };
assert_eq!(sized, un_sized);
assert_eq!(sized_offset, un_sized_offset);
}

View File

@ -0,0 +1,21 @@
// run-pass
use std::ptr::addr_of;
// When the unsized tail is a `dyn Trait`, its alignments is only dynamically known. This means the
// packed(2) needs to be applied at runtime: the actual alignment of the field is `min(2,
// usual_alignment)`. Here we check that we do this right by comparing size, alignment, and field
// offset before and after unsizing.
fn main() {
#[repr(C, packed(2))]
struct Packed<T: ?Sized>(u8, core::mem::ManuallyDrop<T>);
let p = Packed(0, core::mem::ManuallyDrop::new(1));
let p: &Packed<usize> = &p;
let sized = (core::mem::size_of_val(p), core::mem::align_of_val(p));
let sized_offset = unsafe { addr_of!(p.1).cast::<u8>().offset_from(addr_of!(p.0)) };
let p: &Packed<dyn Send> = p;
let un_sized = (core::mem::size_of_val(p), core::mem::align_of_val(p));
let un_sized_offset = unsafe { addr_of!(p.1).cast::<u8>().offset_from(addr_of!(p.0)) };
assert_eq!(sized, un_sized);
assert_eq!(sized_offset, un_sized_offset);
}