forked from rust-lang/rust
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathallocation.rs
1338 lines (1202 loc) · 52.9 KB
/
allocation.rs
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
//! The virtual memory representation of the MIR interpreter.
use std::borrow::Cow;
use std::convert::{TryFrom, TryInto};
use std::fmt;
use std::hash;
use std::iter;
use std::ops::{Deref, Range};
use std::ptr;
use std::mem::MaybeUninit;
use rustc_ast::Mutability;
use rustc_data_structures::intern::Interned;
use rustc_data_structures::sorted_map::SortedMap;
use rustc_span::DUMMY_SP;
use rustc_target::abi::{Align, HasDataLayout, Size};
use super::{
read_target_uint, write_target_uint, AllocId, InterpError, InterpResult, Pointer, Provenance,
ResourceExhaustionInfo, Scalar, ScalarMaybeUninit, ScalarSizeMismatch, UndefinedBehaviorInfo,
UninitBytesAccess, UnsupportedOpInfo,
};
use crate::ty;
/// This type represents an Allocation in the Miri/CTFE core engine.
///
/// Its public API is rather low-level, working directly with allocation offsets and a custom error
/// type to account for the lack of an AllocId on this level. The Miri/CTFE core engine `memory`
/// module provides higher-level access.
// Note: for performance reasons when interning, some of the `Allocation` fields can be partially
// hashed. (see the `Hash` impl below for more details), so the impl is not derived.
#[derive(Clone, Debug, Eq, PartialEq, PartialOrd, Ord, TyEncodable, TyDecodable)]
#[derive(HashStable)]
pub struct Allocation<Prov = AllocId, Extra = ()> {
/// The actual bytes of the allocation.
/// Note that the bytes of a pointer represent the offset of the pointer.
bytes: Box<[u8]>,
/// Maps from byte addresses to extra data for each pointer.
/// Only the first byte of a pointer is inserted into the map; i.e.,
/// every entry in this map applies to `pointer_size` consecutive bytes starting
/// at the given offset.
relocations: Relocations<Prov>,
/// Denotes which part of this allocation is initialized.
init_mask: InitMask,
/// The alignment of the allocation to detect unaligned reads.
/// (`Align` guarantees that this is a power of two.)
pub align: Align,
/// `true` if the allocation is mutable.
/// Also used by codegen to determine if a static should be put into mutable memory,
/// which happens for `static mut` and `static` with interior mutability.
pub mutability: Mutability,
/// Extra state for the machine.
pub extra: Extra,
}
/// This is the maximum size we will hash at a time, when interning an `Allocation` and its
/// `InitMask`. Note, we hash that amount of bytes twice: at the start, and at the end of a buffer.
/// Used when these two structures are large: we only partially hash the larger fields in that
/// situation. See the comment at the top of their respective `Hash` impl for more details.
const MAX_BYTES_TO_HASH: usize = 64;
/// This is the maximum size (in bytes) for which a buffer will be fully hashed, when interning.
/// Otherwise, it will be partially hashed in 2 slices, requiring at least 2 `MAX_BYTES_TO_HASH`
/// bytes.
const MAX_HASHED_BUFFER_LEN: usize = 2 * MAX_BYTES_TO_HASH;
// Const allocations are only hashed for interning. However, they can be large, making the hashing
// expensive especially since it uses `FxHash`: it's better suited to short keys, not potentially
// big buffers like the actual bytes of allocation. We can partially hash some fields when they're
// large.
impl hash::Hash for Allocation {
fn hash<H: hash::Hasher>(&self, state: &mut H) {
// Partially hash the `bytes` buffer when it is large. To limit collisions with common
// prefixes and suffixes, we hash the length and some slices of the buffer.
let byte_count = self.bytes.len();
if byte_count > MAX_HASHED_BUFFER_LEN {
// Hash the buffer's length.
byte_count.hash(state);
// And its head and tail.
self.bytes[..MAX_BYTES_TO_HASH].hash(state);
self.bytes[byte_count - MAX_BYTES_TO_HASH..].hash(state);
} else {
self.bytes.hash(state);
}
// Hash the other fields as usual.
self.relocations.hash(state);
self.init_mask.hash(state);
self.align.hash(state);
self.mutability.hash(state);
self.extra.hash(state);
}
}
/// Interned types generally have an `Outer` type and an `Inner` type, where
/// `Outer` is a newtype around `Interned<Inner>`, and all the operations are
/// done on `Outer`, because all occurrences are interned. E.g. `Ty` is an
/// outer type and `TyS` is its inner type.
///
/// Here things are different because only const allocations are interned. This
/// means that both the inner type (`Allocation`) and the outer type
/// (`ConstAllocation`) are used quite a bit.
#[derive(Copy, Clone, PartialEq, Eq, PartialOrd, Ord, Hash, HashStable)]
#[rustc_pass_by_value]
pub struct ConstAllocation<'tcx, Prov = AllocId, Extra = ()>(
pub Interned<'tcx, Allocation<Prov, Extra>>,
);
impl<'tcx> fmt::Debug for ConstAllocation<'tcx> {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
// This matches how `Allocation` is printed. We print it like this to
// avoid having to update expected output in a lot of tests.
write!(f, "{:?}", self.inner())
}
}
impl<'tcx, Prov, Extra> ConstAllocation<'tcx, Prov, Extra> {
pub fn inner(self) -> &'tcx Allocation<Prov, Extra> {
self.0.0
}
}
/// We have our own error type that does not know about the `AllocId`; that information
/// is added when converting to `InterpError`.
#[derive(Debug)]
pub enum AllocError {
/// A scalar had the wrong size.
ScalarSizeMismatch(ScalarSizeMismatch),
/// Encountered a pointer where we needed raw bytes.
ReadPointerAsBytes,
/// Partially overwriting a pointer.
PartialPointerOverwrite(Size),
/// Using uninitialized data where it is not allowed.
InvalidUninitBytes(Option<UninitBytesAccess>),
}
pub type AllocResult<T = ()> = Result<T, AllocError>;
impl From<ScalarSizeMismatch> for AllocError {
fn from(s: ScalarSizeMismatch) -> Self {
AllocError::ScalarSizeMismatch(s)
}
}
impl AllocError {
pub fn to_interp_error<'tcx>(self, alloc_id: AllocId) -> InterpError<'tcx> {
use AllocError::*;
match self {
ScalarSizeMismatch(s) => {
InterpError::UndefinedBehavior(UndefinedBehaviorInfo::ScalarSizeMismatch(s))
}
ReadPointerAsBytes => InterpError::Unsupported(UnsupportedOpInfo::ReadPointerAsBytes),
PartialPointerOverwrite(offset) => InterpError::Unsupported(
UnsupportedOpInfo::PartialPointerOverwrite(Pointer::new(alloc_id, offset)),
),
InvalidUninitBytes(info) => InterpError::UndefinedBehavior(
UndefinedBehaviorInfo::InvalidUninitBytes(info.map(|b| (alloc_id, b))),
),
}
}
}
/// The information that makes up a memory access: offset and size.
#[derive(Copy, Clone)]
pub struct AllocRange {
pub start: Size,
pub size: Size,
}
impl fmt::Debug for AllocRange {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
write!(f, "[{:#x}..{:#x}]", self.start.bytes(), self.end().bytes())
}
}
/// Free-starting constructor for less syntactic overhead.
#[inline(always)]
pub fn alloc_range(start: Size, size: Size) -> AllocRange {
AllocRange { start, size }
}
impl AllocRange {
#[inline]
pub fn from(r: Range<Size>) -> Self {
alloc_range(r.start, r.end - r.start) // `Size` subtraction (overflow-checked)
}
#[inline(always)]
pub fn end(self) -> Size {
self.start + self.size // This does overflow checking.
}
/// Returns the `subrange` within this range; panics if it is not a subrange.
#[inline]
pub fn subrange(self, subrange: AllocRange) -> AllocRange {
let sub_start = self.start + subrange.start;
let range = alloc_range(sub_start, subrange.size);
assert!(range.end() <= self.end(), "access outside the bounds for given AllocRange");
range
}
}
// The constructors are all without extra; the extra gets added by a machine hook later.
impl<Prov> Allocation<Prov> {
/// Creates an allocation initialized by the given bytes
pub fn from_bytes<'a>(
slice: impl Into<Cow<'a, [u8]>>,
align: Align,
mutability: Mutability,
) -> Self {
let slice: Cow<'a, [u8]> = slice.into();
let size = Size::from_bytes(slice.len());
let align_usize: usize = align.bytes().try_into().unwrap();
let layout = std::alloc::Layout::from_size_align(slice.len(), align_usize).unwrap();
let bytes = unsafe {
let buf = std::alloc::alloc(layout);
let mut uninit_bytes = Vec::from_raw_parts(buf as *mut MaybeUninit<u8>, slice.len(), slice.len());
let mut boxed = Box::<[MaybeUninit<u8>]>::from_raw(&mut *uninit_bytes);
MaybeUninit::write_slice(&mut boxed, &slice);
boxed.assume_init()
};
assert!(bytes.as_ptr() as u64 % align.bytes() == 0);
Self {
bytes,
relocations: Relocations::new(),
init_mask: InitMask::new(size, true),
align,
mutability,
extra: (),
}
}
pub fn from_bytes_byte_aligned_immutable<'a>(slice: impl Into<Cow<'a, [u8]>>) -> Self {
Allocation::from_bytes(slice, Align::ONE, Mutability::Not)
}
/// Try to create an Allocation of `size` bytes, failing if there is not enough memory
/// available to the compiler to do so.
///
/// If `panic_on_fail` is true, this will never return `Err`.
pub fn uninit<'tcx>(size: Size, align: Align, panic_on_fail: bool) -> InterpResult<'tcx, Self> {
let align_usize: usize = align.bytes().try_into().unwrap();
let layout = std::alloc::Layout::from_size_align(size.bytes_usize(), align_usize).unwrap();
let vec_align = unsafe {
// https://doc.rust-lang.org/nightly/std/alloc/trait.GlobalAlloc.html#tymethod.alloc
// std::alloc::alloc returns null to indicate an allocation failure:
// "Returning a null pointer indicates that either memory is exhausted
// or layout does not meet this allocator’s size or alignment constraints."
let buf = std::alloc::alloc(layout);
// Handle allocation failure.
if buf.is_null() {
if panic_on_fail {
panic!("Allocation::uninit called with panic_on_fail had allocation failure")
}
ty::tls::with(|tcx| {
tcx.sess.delay_span_bug(DUMMY_SP, "exhausted memory during interpretation")
});
Err(InterpError::ResourceExhaustion(ResourceExhaustionInfo::MemoryExhausted))?
}
Vec::from_raw_parts(buf as *mut u8, size.bytes_usize(), size.bytes_usize())
};
let bytes = vec_align.into_boxed_slice();
assert!(bytes.as_ptr() as u64 % align.bytes() == 0);
Ok(Allocation {
bytes,
relocations: Relocations::new(),
init_mask: InitMask::new(size, false),
align,
mutability: Mutability::Mut,
extra: (),
})
}
}
impl Allocation {
/// Adjust allocation from the ones in tcx to a custom Machine instance
/// with a different Provenance and Extra type.
pub fn adjust_from_tcx<Prov, Extra, Err>(
self,
cx: &impl HasDataLayout,
extra: Extra,
mut adjust_ptr: impl FnMut(Pointer<AllocId>) -> Result<Pointer<Prov>, Err>,
) -> Result<Allocation<Prov, Extra>, Err> {
// Compute new pointer provenance, which also adjusts the bytes.
// Realign the pointer
let align_usize: usize = self.align.bytes().try_into().unwrap();
let layout = std::alloc::Layout::from_size_align(self.bytes.len(), align_usize).unwrap();
let mut bytes = unsafe {
let buf = std::alloc::alloc(layout);
let mut uninit_bytes = Vec::from_raw_parts(buf as *mut MaybeUninit<u8>, self.bytes.len(), self.bytes.len());
let mut boxed = Box::<[MaybeUninit<u8>]>::from_raw(&mut *uninit_bytes);
MaybeUninit::write_slice(&mut boxed, &self.bytes);
boxed.assume_init()
};
assert!(bytes.as_ptr() as usize % align_usize == 0);
let mut new_relocations = Vec::with_capacity(self.relocations.0.len());
let ptr_size = cx.data_layout().pointer_size.bytes_usize();
let endian = cx.data_layout().endian;
for &(offset, alloc_id) in self.relocations.iter() {
let idx = offset.bytes_usize();
let ptr_bytes = &mut bytes[idx..idx + ptr_size];
let bits = read_target_uint(endian, ptr_bytes).unwrap();
let (ptr_prov, ptr_offset) =
adjust_ptr(Pointer::new(alloc_id, Size::from_bytes(bits)))?.into_parts();
write_target_uint(endian, ptr_bytes, ptr_offset.bytes().into()).unwrap();
new_relocations.push((offset, ptr_prov));
}
assert!(bytes.as_ptr() as u64 % self.align.bytes() == 0);
// Create allocation.
Ok(Allocation {
bytes,
relocations: Relocations::from_presorted(new_relocations),
init_mask: self.init_mask,
align: self.align,
mutability: self.mutability,
extra,
})
}
}
/// Raw accessors. Provide access to otherwise private bytes.
impl<Prov, Extra> Allocation<Prov, Extra> {
pub fn len(&self) -> usize {
self.bytes.len()
}
pub fn size(&self) -> Size {
Size::from_bytes(self.len())
}
/// Looks at a slice which may describe uninitialized bytes or describe a relocation. This differs
/// from `get_bytes_with_uninit_and_ptr` in that it does no relocation checks (even on the
/// edges) at all.
/// This must not be used for reads affecting the interpreter execution.
pub fn inspect_with_uninit_and_ptr_outside_interpreter(&self, range: Range<usize>) -> &[u8] {
&self.bytes[range]
}
/// Returns the mask indicating which bytes are initialized.
pub fn init_mask(&self) -> &InitMask {
&self.init_mask
}
/// Returns the relocation list.
pub fn relocations(&self) -> &Relocations<Prov> {
&self.relocations
}
}
/// Byte accessors.
impl<Prov: Provenance, Extra> Allocation<Prov, Extra> {
/// Get the pointer of the [u8] of bytes.
pub fn get_bytes_addr(&self) -> Size {
Size::from_bytes(self.bytes.as_ptr() as u64)
}
/// This is the entirely abstraction-violating way to just grab the raw bytes without
/// caring about relocations. It just deduplicates some code between `read_scalar`
/// and `get_bytes_internal`.
fn get_bytes_even_more_internal(&self, range: AllocRange) -> &[u8] {
&self.bytes[range.start.bytes_usize()..range.end().bytes_usize()]
}
/// The last argument controls whether we error out when there are uninitialized or pointer
/// bytes. However, we *always* error when there are relocations overlapping the edges of the
/// range.
///
/// You should never call this, call `get_bytes` or `get_bytes_with_uninit_and_ptr` instead,
///
/// This function also guarantees that the resulting pointer will remain stable
/// even when new allocations are pushed to the `HashMap`. `mem_copy_repeatedly` relies
/// on that.
///
/// It is the caller's responsibility to check bounds and alignment beforehand.
fn get_bytes_internal(
&self,
cx: &impl HasDataLayout,
range: AllocRange,
check_init_and_ptr: bool,
) -> AllocResult<&[u8]> {
if check_init_and_ptr {
self.check_init(range)?;
self.check_relocations(cx, range)?;
} else {
// We still don't want relocations on the *edges*.
self.check_relocation_edges(cx, range)?;
}
Ok(self.get_bytes_even_more_internal(range))
}
/// Checks that these bytes are initialized and not pointer bytes, and then return them
/// as a slice.
///
/// It is the caller's responsibility to check bounds and alignment beforehand.
/// Most likely, you want to use the `PlaceTy` and `OperandTy`-based methods
/// on `InterpCx` instead.
#[inline]
pub fn get_bytes(&self, cx: &impl HasDataLayout, range: AllocRange) -> AllocResult<&[u8]> {
self.get_bytes_internal(cx, range, true)
}
/// It is the caller's responsibility to handle uninitialized and pointer bytes.
/// However, this still checks that there are no relocations on the *edges*.
///
/// It is the caller's responsibility to check bounds and alignment beforehand.
#[inline]
pub fn get_bytes_with_uninit_and_ptr(
&self,
cx: &impl HasDataLayout,
range: AllocRange,
) -> AllocResult<&[u8]> {
self.get_bytes_internal(cx, range, false)
}
/// Just calling this already marks everything as defined and removes relocations,
/// so be sure to actually put data there!
///
/// It is the caller's responsibility to check bounds and alignment beforehand.
/// Most likely, you want to use the `PlaceTy` and `OperandTy`-based methods
/// on `InterpCx` instead.
pub fn get_bytes_mut(
&mut self,
cx: &impl HasDataLayout,
range: AllocRange,
) -> AllocResult<&mut [u8]> {
self.mark_init(range, true);
self.clear_relocations(cx, range)?;
Ok(&mut self.bytes[range.start.bytes_usize()..range.end().bytes_usize()])
}
/// A raw pointer variant of `get_bytes_mut` that avoids invalidating existing aliases into this memory.
pub fn get_bytes_mut_ptr(
&mut self,
cx: &impl HasDataLayout,
range: AllocRange,
) -> AllocResult<*mut [u8]> {
self.mark_init(range, true);
self.clear_relocations(cx, range)?;
assert!(range.end().bytes_usize() <= self.bytes.len()); // need to do our own bounds-check
let begin_ptr = self.bytes.as_mut_ptr().wrapping_add(range.start.bytes_usize());
let len = range.end().bytes_usize() - range.start.bytes_usize();
Ok(ptr::slice_from_raw_parts_mut(begin_ptr, len))
}
}
/// Reading and writing.
impl<Prov: Provenance, Extra> Allocation<Prov, Extra> {
/// Validates that `ptr.offset` and `ptr.offset + size` do not point to the middle of a
/// relocation. If `allow_uninit`/`allow_ptr` is `false`, also enforces that the memory in the
/// given range contains no uninitialized bytes/relocations.
pub fn check_bytes(
&self,
cx: &impl HasDataLayout,
range: AllocRange,
allow_uninit: bool,
allow_ptr: bool,
) -> AllocResult {
// Check bounds and relocations on the edges.
self.get_bytes_with_uninit_and_ptr(cx, range)?;
// Check uninit and ptr.
if !allow_uninit {
self.check_init(range)?;
}
if !allow_ptr {
self.check_relocations(cx, range)?;
}
Ok(())
}
/// Reads a *non-ZST* scalar.
///
/// If `read_provenance` is `true`, this will also read provenance; otherwise (if the machine
/// supports that) provenance is entirely ignored.
///
/// ZSTs can't be read because in order to obtain a `Pointer`, we need to check
/// for ZSTness anyway due to integer pointers being valid for ZSTs.
///
/// It is the caller's responsibility to check bounds and alignment beforehand.
/// Most likely, you want to call `InterpCx::read_scalar` instead of this method.
pub fn read_scalar(
&self,
cx: &impl HasDataLayout,
range: AllocRange,
read_provenance: bool,
) -> AllocResult<ScalarMaybeUninit<Prov>> {
if read_provenance {
assert_eq!(range.size, cx.data_layout().pointer_size);
}
// First and foremost, if anything is uninit, bail.
if self.is_init(range).is_err() {
// This inflates uninitialized bytes to the entire scalar, even if only a few
// bytes are uninitialized.
return Ok(ScalarMaybeUninit::Uninit);
}
// If we are doing a pointer read, and there is a relocation exactly where we
// are reading, then we can put data and relocation back together and return that.
if read_provenance && let Some(&prov) = self.relocations.get(&range.start) {
// We already checked init and relocations, so we can use this function.
let bytes = self.get_bytes_even_more_internal(range);
let bits = read_target_uint(cx.data_layout().endian, bytes).unwrap();
let ptr = Pointer::new(prov, Size::from_bytes(bits));
return Ok(ScalarMaybeUninit::from_pointer(ptr, cx));
}
// If we are *not* reading a pointer, and we can just ignore relocations,
// then do exactly that.
if !read_provenance && Prov::OFFSET_IS_ADDR {
// We just strip provenance.
let bytes = self.get_bytes_even_more_internal(range);
let bits = read_target_uint(cx.data_layout().endian, bytes).unwrap();
return Ok(ScalarMaybeUninit::Scalar(Scalar::from_uint(bits, range.size)));
}
// It's complicated. Better make sure there is no provenance anywhere.
// FIXME: If !OFFSET_IS_ADDR, this is the best we can do. But if OFFSET_IS_ADDR, then
// `read_pointer` is true and we ideally would distinguish the following two cases:
// - The entire `range` is covered by 2 relocations for the same provenance.
// Then we should return a pointer with that provenance.
// - The range has inhomogeneous provenance. Then we should return just the
// underlying bits.
let bytes = self.get_bytes(cx, range)?;
let bits = read_target_uint(cx.data_layout().endian, bytes).unwrap();
Ok(ScalarMaybeUninit::Scalar(Scalar::from_uint(bits, range.size)))
}
/// Writes a *non-ZST* scalar.
///
/// ZSTs can't be read because in order to obtain a `Pointer`, we need to check
/// for ZSTness anyway due to integer pointers being valid for ZSTs.
///
/// It is the caller's responsibility to check bounds and alignment beforehand.
/// Most likely, you want to call `InterpCx::write_scalar` instead of this method.
#[instrument(skip(self, cx), level = "debug")]
pub fn write_scalar(
&mut self,
cx: &impl HasDataLayout,
range: AllocRange,
val: ScalarMaybeUninit<Prov>,
) -> AllocResult {
assert!(self.mutability == Mutability::Mut);
let val = match val {
ScalarMaybeUninit::Scalar(scalar) => scalar,
ScalarMaybeUninit::Uninit => {
return self.write_uninit(cx, range);
}
};
// `to_bits_or_ptr_internal` is the right method because we just want to store this data
// as-is into memory.
let (bytes, provenance) = match val.to_bits_or_ptr_internal(range.size)? {
Err(val) => {
let (provenance, offset) = val.into_parts();
(u128::from(offset.bytes()), Some(provenance))
}
Ok(data) => (data, None),
};
let endian = cx.data_layout().endian;
let dst = self.get_bytes_mut(cx, range)?;
write_target_uint(endian, dst, bytes).unwrap();
// See if we have to also write a relocation.
if let Some(provenance) = provenance {
self.relocations.0.insert(range.start, provenance);
}
Ok(())
}
/// Write "uninit" to the given memory range.
pub fn write_uninit(&mut self, cx: &impl HasDataLayout, range: AllocRange) -> AllocResult {
self.mark_init(range, false);
self.clear_relocations(cx, range)?;
return Ok(());
}
}
/// Relocations.
impl<Prov: Copy, Extra> Allocation<Prov, Extra> {
/// Returns all relocations overlapping with the given pointer-offset pair.
fn get_relocations(&self, cx: &impl HasDataLayout, range: AllocRange) -> &[(Size, Prov)] {
// We have to go back `pointer_size - 1` bytes, as that one would still overlap with
// the beginning of this range.
let start = range.start.bytes().saturating_sub(cx.data_layout().pointer_size.bytes() - 1);
self.relocations.range(Size::from_bytes(start)..range.end())
}
/// Returns whether this allocation has relocations overlapping with the given range.
///
/// Note: this function exists to allow `get_relocations` to be private, in order to somewhat
/// limit access to relocations outside of the `Allocation` abstraction.
///
pub fn has_relocations(&self, cx: &impl HasDataLayout, range: AllocRange) -> bool {
!self.get_relocations(cx, range).is_empty()
}
/// Checks that there are no relocations overlapping with the given range.
#[inline(always)]
fn check_relocations(&self, cx: &impl HasDataLayout, range: AllocRange) -> AllocResult {
if self.has_relocations(cx, range) { Err(AllocError::ReadPointerAsBytes) } else { Ok(()) }
}
/// Removes all relocations inside the given range.
/// If there are relocations overlapping with the edges, they
/// are removed as well *and* the bytes they cover are marked as
/// uninitialized. This is a somewhat odd "spooky action at a distance",
/// but it allows strictly more code to run than if we would just error
/// immediately in that case.
fn clear_relocations(&mut self, cx: &impl HasDataLayout, range: AllocRange) -> AllocResult
where
Prov: Provenance,
{
// Find the start and end of the given range and its outermost relocations.
let (first, last) = {
// Find all relocations overlapping the given range.
let relocations = self.get_relocations(cx, range);
if relocations.is_empty() {
return Ok(());
}
(
relocations.first().unwrap().0,
relocations.last().unwrap().0 + cx.data_layout().pointer_size,
)
};
let start = range.start;
let end = range.end();
// We need to handle clearing the relocations from parts of a pointer.
// FIXME: Miri should preserve partial relocations; see
// https://github.com/rust-lang/miri/issues/2181.
if first < start {
if Prov::ERR_ON_PARTIAL_PTR_OVERWRITE {
return Err(AllocError::PartialPointerOverwrite(first));
}
warn!(
"Partial pointer overwrite! De-initializing memory at offsets {first:?}..{start:?}."
);
self.init_mask.set_range(first, start, false);
}
if last > end {
if Prov::ERR_ON_PARTIAL_PTR_OVERWRITE {
return Err(AllocError::PartialPointerOverwrite(
last - cx.data_layout().pointer_size,
));
}
warn!(
"Partial pointer overwrite! De-initializing memory at offsets {end:?}..{last:?}."
);
self.init_mask.set_range(end, last, false);
}
// Forget all the relocations.
// Since relocations do not overlap, we know that removing until `last` (exclusive) is fine,
// i.e., this will not remove any other relocations just after the ones we care about.
self.relocations.0.remove_range(first..last);
Ok(())
}
/// Errors if there are relocations overlapping with the edges of the
/// given memory range.
#[inline]
fn check_relocation_edges(&self, cx: &impl HasDataLayout, range: AllocRange) -> AllocResult {
self.check_relocations(cx, alloc_range(range.start, Size::ZERO))?;
self.check_relocations(cx, alloc_range(range.end(), Size::ZERO))?;
Ok(())
}
}
/// "Relocations" stores the provenance information of pointers stored in memory.
#[derive(Clone, PartialEq, Eq, PartialOrd, Ord, Hash, Debug, TyEncodable, TyDecodable)]
pub struct Relocations<Prov = AllocId>(SortedMap<Size, Prov>);
impl<Prov> Relocations<Prov> {
pub fn new() -> Self {
Relocations(SortedMap::new())
}
// The caller must guarantee that the given relocations are already sorted
// by address and contain no duplicates.
pub fn from_presorted(r: Vec<(Size, Prov)>) -> Self {
Relocations(SortedMap::from_presorted_elements(r))
}
}
impl<Prov> Deref for Relocations<Prov> {
type Target = SortedMap<Size, Prov>;
fn deref(&self) -> &Self::Target {
&self.0
}
}
/// A partial, owned list of relocations to transfer into another allocation.
///
/// Offsets are already adjusted to the destination allocation.
pub struct AllocationRelocations<Prov> {
dest_relocations: Vec<(Size, Prov)>,
}
impl<Prov: Copy, Extra> Allocation<Prov, Extra> {
pub fn prepare_relocation_copy(
&self,
cx: &impl HasDataLayout,
src: AllocRange,
dest: Size,
count: u64,
) -> AllocationRelocations<Prov> {
let relocations = self.get_relocations(cx, src);
if relocations.is_empty() {
return AllocationRelocations { dest_relocations: Vec::new() };
}
let size = src.size;
let mut new_relocations = Vec::with_capacity(relocations.len() * (count as usize));
// If `count` is large, this is rather wasteful -- we are allocating a big array here, which
// is mostly filled with redundant information since it's just N copies of the same `Prov`s
// at slightly adjusted offsets. The reason we do this is so that in `mark_relocation_range`
// we can use `insert_presorted`. That wouldn't work with an `Iterator` that just produces
// the right sequence of relocations for all N copies.
for i in 0..count {
new_relocations.extend(relocations.iter().map(|&(offset, reloc)| {
// compute offset for current repetition
let dest_offset = dest + size * i; // `Size` operations
(
// shift offsets from source allocation to destination allocation
(offset + dest_offset) - src.start, // `Size` operations
reloc,
)
}));
}
AllocationRelocations { dest_relocations: new_relocations }
}
/// Applies a relocation copy.
/// The affected range, as defined in the parameters to `prepare_relocation_copy` is expected
/// to be clear of relocations.
///
/// This is dangerous to use as it can violate internal `Allocation` invariants!
/// It only exists to support an efficient implementation of `mem_copy_repeatedly`.
pub fn mark_relocation_range(&mut self, relocations: AllocationRelocations<Prov>) {
self.relocations.0.insert_presorted(relocations.dest_relocations);
}
}
////////////////////////////////////////////////////////////////////////////////
// Uninitialized byte tracking
////////////////////////////////////////////////////////////////////////////////
type Block = u64;
/// A bitmask where each bit refers to the byte with the same index. If the bit is `true`, the byte
/// is initialized. If it is `false` the byte is uninitialized.
// Note: for performance reasons when interning, some of the `InitMask` fields can be partially
// hashed. (see the `Hash` impl below for more details), so the impl is not derived.
#[derive(Clone, Debug, Eq, PartialEq, PartialOrd, Ord, TyEncodable, TyDecodable)]
#[derive(HashStable)]
pub struct InitMask {
blocks: Vec<Block>,
len: Size,
}
// Const allocations are only hashed for interning. However, they can be large, making the hashing
// expensive especially since it uses `FxHash`: it's better suited to short keys, not potentially
// big buffers like the allocation's init mask. We can partially hash some fields when they're
// large.
impl hash::Hash for InitMask {
fn hash<H: hash::Hasher>(&self, state: &mut H) {
const MAX_BLOCKS_TO_HASH: usize = MAX_BYTES_TO_HASH / std::mem::size_of::<Block>();
const MAX_BLOCKS_LEN: usize = MAX_HASHED_BUFFER_LEN / std::mem::size_of::<Block>();
// Partially hash the `blocks` buffer when it is large. To limit collisions with common
// prefixes and suffixes, we hash the length and some slices of the buffer.
let block_count = self.blocks.len();
if block_count > MAX_BLOCKS_LEN {
// Hash the buffer's length.
block_count.hash(state);
// And its head and tail.
self.blocks[..MAX_BLOCKS_TO_HASH].hash(state);
self.blocks[block_count - MAX_BLOCKS_TO_HASH..].hash(state);
} else {
self.blocks.hash(state);
}
// Hash the other fields as usual.
self.len.hash(state);
}
}
impl InitMask {
pub const BLOCK_SIZE: u64 = 64;
#[inline]
fn bit_index(bits: Size) -> (usize, usize) {
// BLOCK_SIZE is the number of bits that can fit in a `Block`.
// Each bit in a `Block` represents the initialization state of one byte of an allocation,
// so we use `.bytes()` here.
let bits = bits.bytes();
let a = bits / InitMask::BLOCK_SIZE;
let b = bits % InitMask::BLOCK_SIZE;
(usize::try_from(a).unwrap(), usize::try_from(b).unwrap())
}
#[inline]
fn size_from_bit_index(block: impl TryInto<u64>, bit: impl TryInto<u64>) -> Size {
let block = block.try_into().ok().unwrap();
let bit = bit.try_into().ok().unwrap();
Size::from_bytes(block * InitMask::BLOCK_SIZE + bit)
}
pub fn new(size: Size, state: bool) -> Self {
let mut m = InitMask { blocks: vec![], len: Size::ZERO };
m.grow(size, state);
m
}
pub fn set_range(&mut self, start: Size, end: Size, new_state: bool) {
let len = self.len;
if end > len {
self.grow(end - len, new_state);
}
self.set_range_inbounds(start, end, new_state);
}
pub fn set_range_inbounds(&mut self, start: Size, end: Size, new_state: bool) {
let (blocka, bita) = Self::bit_index(start);
let (blockb, bitb) = Self::bit_index(end);
if blocka == blockb {
// First set all bits except the first `bita`,
// then unset the last `64 - bitb` bits.
let range = if bitb == 0 {
u64::MAX << bita
} else {
(u64::MAX << bita) & (u64::MAX >> (64 - bitb))
};
if new_state {
self.blocks[blocka] |= range;
} else {
self.blocks[blocka] &= !range;
}
return;
}
// across block boundaries
if new_state {
// Set `bita..64` to `1`.
self.blocks[blocka] |= u64::MAX << bita;
// Set `0..bitb` to `1`.
if bitb != 0 {
self.blocks[blockb] |= u64::MAX >> (64 - bitb);
}
// Fill in all the other blocks (much faster than one bit at a time).
for block in (blocka + 1)..blockb {
self.blocks[block] = u64::MAX;
}
} else {
// Set `bita..64` to `0`.
self.blocks[blocka] &= !(u64::MAX << bita);
// Set `0..bitb` to `0`.
if bitb != 0 {
self.blocks[blockb] &= !(u64::MAX >> (64 - bitb));
}
// Fill in all the other blocks (much faster than one bit at a time).
for block in (blocka + 1)..blockb {
self.blocks[block] = 0;
}
}
}
#[inline]
pub fn get(&self, i: Size) -> bool {
let (block, bit) = Self::bit_index(i);
(self.blocks[block] & (1 << bit)) != 0
}
#[inline]
pub fn set(&mut self, i: Size, new_state: bool) {
let (block, bit) = Self::bit_index(i);
self.set_bit(block, bit, new_state);
}
#[inline]
fn set_bit(&mut self, block: usize, bit: usize, new_state: bool) {
if new_state {
self.blocks[block] |= 1 << bit;
} else {
self.blocks[block] &= !(1 << bit);
}
}
pub fn grow(&mut self, amount: Size, new_state: bool) {
if amount.bytes() == 0 {
return;
}
let unused_trailing_bits =
u64::try_from(self.blocks.len()).unwrap() * Self::BLOCK_SIZE - self.len.bytes();
if amount.bytes() > unused_trailing_bits {
let additional_blocks = amount.bytes() / Self::BLOCK_SIZE + 1;
self.blocks.extend(
// FIXME(oli-obk): optimize this by repeating `new_state as Block`.
iter::repeat(0).take(usize::try_from(additional_blocks).unwrap()),
);
}
let start = self.len;
self.len += amount;
self.set_range_inbounds(start, start + amount, new_state); // `Size` operation
}
/// Returns the index of the first bit in `start..end` (end-exclusive) that is equal to is_init.
fn find_bit(&self, start: Size, end: Size, is_init: bool) -> Option<Size> {
/// A fast implementation of `find_bit`,
/// which skips over an entire block at a time if it's all 0s (resp. 1s),
/// and finds the first 1 (resp. 0) bit inside a block using `trailing_zeros` instead of a loop.
///
/// Note that all examples below are written with 8 (instead of 64) bit blocks for simplicity,
/// and with the least significant bit (and lowest block) first:
/// ```text
/// 00000000|00000000
/// ^ ^ ^ ^
/// index: 0 7 8 15
/// ```
/// Also, if not stated, assume that `is_init = true`, that is, we are searching for the first 1 bit.
fn find_bit_fast(
init_mask: &InitMask,
start: Size,
end: Size,
is_init: bool,
) -> Option<Size> {
/// Search one block, returning the index of the first bit equal to `is_init`.
fn search_block(
bits: Block,
block: usize,
start_bit: usize,
is_init: bool,
) -> Option<Size> {
// For the following examples, assume this function was called with:
// bits = 0b00111011
// start_bit = 3
// is_init = false
// Note that, for the examples in this function, the most significant bit is written first,
// which is backwards compared to the comments in `find_bit`/`find_bit_fast`.
// Invert bits so we're always looking for the first set bit.
// ! 0b00111011
// bits = 0b11000100
let bits = if is_init { bits } else { !bits };
// Mask off unused start bits.
// 0b11000100
// & 0b11111000
// bits = 0b11000000
let bits = bits & (!0 << start_bit);
// Find set bit, if any.
// bit = trailing_zeros(0b11000000)
// bit = 6
if bits == 0 {
None
} else {
let bit = bits.trailing_zeros();
Some(InitMask::size_from_bit_index(block, bit))
}
}
if start >= end {
return None;
}
// Convert `start` and `end` to block indexes and bit indexes within each block.
// We must convert `end` to an inclusive bound to handle block boundaries correctly.
//
// For example:
//
// (a) 00000000|00000000 (b) 00000000|
// ^~~~~~~~~~~^ ^~~~~~~~~^
// start end start end
//
// In both cases, the block index of `end` is 1.
// But we do want to search block 1 in (a), and we don't in (b).
//
// We subtract 1 from both end positions to make them inclusive:
//
// (a) 00000000|00000000 (b) 00000000|
// ^~~~~~~~~~^ ^~~~~~~^
// start end_inclusive start end_inclusive
//
// For (a), the block index of `end_inclusive` is 1, and for (b), it's 0.
// This provides the desired behavior of searching blocks 0 and 1 for (a),