1//===-- tsan_rtl.h ----------------------------------------------*- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file is a part of ThreadSanitizer (TSan), a race detector.
10//
11// Main internal TSan header file.
12//
13// Ground rules:
14// - C++ run-time should not be used (static CTORs, RTTI, exceptions, static
15// function-scope locals)
16// - All functions/classes/etc reside in namespace __tsan, except for those
17// declared in tsan_interface.h.
18// - Platform-specific files should be used instead of ifdefs (*).
19// - No system headers included in header files (*).
20// - Platform specific headres included only into platform-specific files (*).
21//
22// (*) Except when inlining is critical for performance.
23//===----------------------------------------------------------------------===//
24
25#ifndef TSAN_RTL_H
26#define TSAN_RTL_H
27
28#include "sanitizer_common/sanitizer_allocator.h"
29#include "sanitizer_common/sanitizer_allocator_internal.h"
30#include "sanitizer_common/sanitizer_asm.h"
31#include "sanitizer_common/sanitizer_common.h"
32#include "sanitizer_common/sanitizer_deadlock_detector_interface.h"
33#include "sanitizer_common/sanitizer_libignore.h"
34#include "sanitizer_common/sanitizer_suppressions.h"
35#include "sanitizer_common/sanitizer_thread_registry.h"
36#include "sanitizer_common/sanitizer_vector.h"
37#include "tsan_clock.h"
38#include "tsan_defs.h"
39#include "tsan_flags.h"
40#include "tsan_mman.h"
41#include "tsan_sync.h"
42#include "tsan_trace.h"
43#include "tsan_report.h"
44#include "tsan_platform.h"
45#include "tsan_mutexset.h"
46#include "tsan_ignoreset.h"
47#include "tsan_stack_trace.h"
48
49#if SANITIZER_WORDSIZE != 64
50# error "ThreadSanitizer is supported only on 64-bit platforms"
51#endif
52
53namespace __tsan {
54
55#if !SANITIZER_GO
56struct MapUnmapCallback;
57#if defined(__mips64) || defined(__aarch64__) || defined(__powerpc__)
58
59struct AP32 {
60 static const uptr kSpaceBeg = 0;
61 static const u64 kSpaceSize = SANITIZER_MMAP_RANGE_SIZE;
62 static const uptr kMetadataSize = 0;
63 typedef __sanitizer::CompactSizeClassMap SizeClassMap;
64 static const uptr kRegionSizeLog = 20;
65 using AddressSpaceView = LocalAddressSpaceView;
66 typedef __tsan::MapUnmapCallback MapUnmapCallback;
67 static const uptr kFlags = 0;
68};
69typedef SizeClassAllocator32<AP32> PrimaryAllocator;
70#else
71struct AP64 { // Allocator64 parameters. Deliberately using a short name.
72 static const uptr kSpaceBeg = Mapping::kHeapMemBeg;
73 static const uptr kSpaceSize = Mapping::kHeapMemEnd - Mapping::kHeapMemBeg;
74 static const uptr kMetadataSize = 0;
75 typedef DefaultSizeClassMap SizeClassMap;
76 typedef __tsan::MapUnmapCallback MapUnmapCallback;
77 static const uptr kFlags = 0;
78 using AddressSpaceView = LocalAddressSpaceView;
79};
80typedef SizeClassAllocator64<AP64> PrimaryAllocator;
81#endif
82typedef CombinedAllocator<PrimaryAllocator> Allocator;
83typedef Allocator::AllocatorCache AllocatorCache;
84Allocator *allocator();
85#endif
86
87void TsanCheckFailed(const char *file, int line, const char *cond,
88 u64 v1, u64 v2);
89
90const u64 kShadowRodata = (u64)-1; // .rodata shadow marker
91
92// FastState (from most significant bit):
93// ignore : 1
94// tid : kTidBits
95// unused : -
96// history_size : 3
97// epoch : kClkBits
98class FastState {
99 public:
100 FastState(u64 tid, u64 epoch) {
101 x_ = tid << kTidShift;
102 x_ |= epoch;
103 DCHECK_EQ(tid, this->tid());
104 DCHECK_EQ(epoch, this->epoch());
105 DCHECK_EQ(GetIgnoreBit(), false);
106 }
107
108 explicit FastState(u64 x)
109 : x_(x) {
110 }
111
112 u64 raw() const {
113 return x_;
114 }
115
116 u64 tid() const {
117 u64 res = (x_ & ~kIgnoreBit) >> kTidShift;
118 return res;
119 }
120
121 u64 TidWithIgnore() const {
122 u64 res = x_ >> kTidShift;
123 return res;
124 }
125
126 u64 epoch() const {
127 u64 res = x_ & ((1ull << kClkBits) - 1);
128 return res;
129 }
130
131 void IncrementEpoch() {
132 u64 old_epoch = epoch();
133 x_ += 1;
134 DCHECK_EQ(old_epoch + 1, epoch());
135 (void)old_epoch;
136 }
137
138 void SetIgnoreBit() { x_ |= kIgnoreBit; }
139 void ClearIgnoreBit() { x_ &= ~kIgnoreBit; }
140 bool GetIgnoreBit() const { return (s64)x_ < 0; }
141
142 void SetHistorySize(int hs) {
143 CHECK_GE(hs, 0);
144 CHECK_LE(hs, 7);
145 x_ = (x_ & ~(kHistoryMask << kHistoryShift)) | (u64(hs) << kHistoryShift);
146 }
147
148 ALWAYS_INLINE
149 int GetHistorySize() const {
150 return (int)((x_ >> kHistoryShift) & kHistoryMask);
151 }
152
153 void ClearHistorySize() {
154 SetHistorySize(0);
155 }
156
157 ALWAYS_INLINE
158 u64 GetTracePos() const {
159 const int hs = GetHistorySize();
160 // When hs == 0, the trace consists of 2 parts.
161 const u64 mask = (1ull << (kTracePartSizeBits + hs + 1)) - 1;
162 return epoch() & mask;
163 }
164
165 private:
166 friend class Shadow;
167 static const int kTidShift = 64 - kTidBits - 1;
168 static const u64 kIgnoreBit = 1ull << 63;
169 static const u64 kFreedBit = 1ull << 63;
170 static const u64 kHistoryShift = kClkBits;
171 static const u64 kHistoryMask = 7;
172 u64 x_;
173};
174
175// Shadow (from most significant bit):
176// freed : 1
177// tid : kTidBits
178// is_atomic : 1
179// is_read : 1
180// size_log : 2
181// addr0 : 3
182// epoch : kClkBits
183class Shadow : public FastState {
184 public:
185 explicit Shadow(u64 x)
186 : FastState(x) {
187 }
188
189 explicit Shadow(const FastState &s)
190 : FastState(s.x_) {
191 ClearHistorySize();
192 }
193
194 void SetAddr0AndSizeLog(u64 addr0, unsigned kAccessSizeLog) {
195 DCHECK_EQ((x_ >> kClkBits) & 31, 0);
196 DCHECK_LE(addr0, 7);
197 DCHECK_LE(kAccessSizeLog, 3);
198 x_ |= ((kAccessSizeLog << 3) | addr0) << kClkBits;
199 DCHECK_EQ(kAccessSizeLog, size_log());
200 DCHECK_EQ(addr0, this->addr0());
201 }
202
203 void SetWrite(unsigned kAccessIsWrite) {
204 DCHECK_EQ(x_ & kReadBit, 0);
205 if (!kAccessIsWrite)
206 x_ |= kReadBit;
207 DCHECK_EQ(kAccessIsWrite, IsWrite());
208 }
209
210 void SetAtomic(bool kIsAtomic) {
211 DCHECK(!IsAtomic());
212 if (kIsAtomic)
213 x_ |= kAtomicBit;
214 DCHECK_EQ(IsAtomic(), kIsAtomic);
215 }
216
217 bool IsAtomic() const {
218 return x_ & kAtomicBit;
219 }
220
221 bool IsZero() const {
222 return x_ == 0;
223 }
224
225 static inline bool TidsAreEqual(const Shadow s1, const Shadow s2) {
226 u64 shifted_xor = (s1.x_ ^ s2.x_) >> kTidShift;
227 DCHECK_EQ(shifted_xor == 0, s1.TidWithIgnore() == s2.TidWithIgnore());
228 return shifted_xor == 0;
229 }
230
231 static ALWAYS_INLINE
232 bool Addr0AndSizeAreEqual(const Shadow s1, const Shadow s2) {
233 u64 masked_xor = ((s1.x_ ^ s2.x_) >> kClkBits) & 31;
234 return masked_xor == 0;
235 }
236
237 static ALWAYS_INLINE bool TwoRangesIntersect(Shadow s1, Shadow s2,
238 unsigned kS2AccessSize) {
239 bool res = false;
240 u64 diff = s1.addr0() - s2.addr0();
241 if ((s64)diff < 0) { // s1.addr0 < s2.addr0
242 // if (s1.addr0() + size1) > s2.addr0()) return true;
243 if (s1.size() > -diff)
244 res = true;
245 } else {
246 // if (s2.addr0() + kS2AccessSize > s1.addr0()) return true;
247 if (kS2AccessSize > diff)
248 res = true;
249 }
250 DCHECK_EQ(res, TwoRangesIntersectSlow(s1, s2));
251 DCHECK_EQ(res, TwoRangesIntersectSlow(s2, s1));
252 return res;
253 }
254
255 u64 ALWAYS_INLINE addr0() const { return (x_ >> kClkBits) & 7; }
256 u64 ALWAYS_INLINE size() const { return 1ull << size_log(); }
257 bool ALWAYS_INLINE IsWrite() const { return !IsRead(); }
258 bool ALWAYS_INLINE IsRead() const { return x_ & kReadBit; }
259
260 // The idea behind the freed bit is as follows.
261 // When the memory is freed (or otherwise unaccessible) we write to the shadow
262 // values with tid/epoch related to the free and the freed bit set.
263 // During memory accesses processing the freed bit is considered
264 // as msb of tid. So any access races with shadow with freed bit set
265 // (it is as if write from a thread with which we never synchronized before).
266 // This allows us to detect accesses to freed memory w/o additional
267 // overheads in memory access processing and at the same time restore
268 // tid/epoch of free.
269 void MarkAsFreed() {
270 x_ |= kFreedBit;
271 }
272
273 bool IsFreed() const {
274 return x_ & kFreedBit;
275 }
276
277 bool GetFreedAndReset() {
278 bool res = x_ & kFreedBit;
279 x_ &= ~kFreedBit;
280 return res;
281 }
282
283 bool ALWAYS_INLINE IsBothReadsOrAtomic(bool kIsWrite, bool kIsAtomic) const {
284 bool v = x_ & ((u64(kIsWrite ^ 1) << kReadShift)
285 | (u64(kIsAtomic) << kAtomicShift));
286 DCHECK_EQ(v, (!IsWrite() && !kIsWrite) || (IsAtomic() && kIsAtomic));
287 return v;
288 }
289
290 bool ALWAYS_INLINE IsRWNotWeaker(bool kIsWrite, bool kIsAtomic) const {
291 bool v = ((x_ >> kReadShift) & 3)
292 <= u64((kIsWrite ^ 1) | (kIsAtomic << 1));
293 DCHECK_EQ(v, (IsAtomic() < kIsAtomic) ||
294 (IsAtomic() == kIsAtomic && !IsWrite() <= !kIsWrite));
295 return v;
296 }
297
298 bool ALWAYS_INLINE IsRWWeakerOrEqual(bool kIsWrite, bool kIsAtomic) const {
299 bool v = ((x_ >> kReadShift) & 3)
300 >= u64((kIsWrite ^ 1) | (kIsAtomic << 1));
301 DCHECK_EQ(v, (IsAtomic() > kIsAtomic) ||
302 (IsAtomic() == kIsAtomic && !IsWrite() >= !kIsWrite));
303 return v;
304 }
305
306 private:
307 static const u64 kReadShift = 5 + kClkBits;
308 static const u64 kReadBit = 1ull << kReadShift;
309 static const u64 kAtomicShift = 6 + kClkBits;
310 static const u64 kAtomicBit = 1ull << kAtomicShift;
311
312 u64 size_log() const { return (x_ >> (3 + kClkBits)) & 3; }
313
314 static bool TwoRangesIntersectSlow(const Shadow s1, const Shadow s2) {
315 if (s1.addr0() == s2.addr0()) return true;
316 if (s1.addr0() < s2.addr0() && s1.addr0() + s1.size() > s2.addr0())
317 return true;
318 if (s2.addr0() < s1.addr0() && s2.addr0() + s2.size() > s1.addr0())
319 return true;
320 return false;
321 }
322};
323
324struct ThreadSignalContext;
325
326struct JmpBuf {
327 uptr sp;
328 int int_signal_send;
329 bool in_blocking_func;
330 uptr in_signal_handler;
331 uptr *shadow_stack_pos;
332};
333
334// A Processor represents a physical thread, or a P for Go.
335// It is used to store internal resources like allocate cache, and does not
336// participate in race-detection logic (invisible to end user).
337// In C++ it is tied to an OS thread just like ThreadState, however ideally
338// it should be tied to a CPU (this way we will have fewer allocator caches).
339// In Go it is tied to a P, so there are significantly fewer Processor's than
340// ThreadState's (which are tied to Gs).
341// A ThreadState must be wired with a Processor to handle events.
342struct Processor {
343 ThreadState *thr; // currently wired thread, or nullptr
344#if !SANITIZER_GO
345 AllocatorCache alloc_cache;
346 InternalAllocatorCache internal_alloc_cache;
347#endif
348 DenseSlabAllocCache block_cache;
349 DenseSlabAllocCache sync_cache;
350 DenseSlabAllocCache clock_cache;
351 DDPhysicalThread *dd_pt;
352};
353
354#if !SANITIZER_GO
355// ScopedGlobalProcessor temporary setups a global processor for the current
356// thread, if it does not have one. Intended for interceptors that can run
357// at the very thread end, when we already destroyed the thread processor.
358struct ScopedGlobalProcessor {
359 ScopedGlobalProcessor();
360 ~ScopedGlobalProcessor();
361};
362#endif
363
364// This struct is stored in TLS.
365struct ThreadState {
366 FastState fast_state;
367 // Synch epoch represents the threads's epoch before the last synchronization
368 // action. It allows to reduce number of shadow state updates.
369 // For example, fast_synch_epoch=100, last write to addr X was at epoch=150,
370 // if we are processing write to X from the same thread at epoch=200,
371 // we do nothing, because both writes happen in the same 'synch epoch'.
372 // That is, if another memory access does not race with the former write,
373 // it does not race with the latter as well.
374 // QUESTION: can we can squeeze this into ThreadState::Fast?
375 // E.g. ThreadState::Fast is a 44-bit, 32 are taken by synch_epoch and 12 are
376 // taken by epoch between synchs.
377 // This way we can save one load from tls.
378 u64 fast_synch_epoch;
379 // Technically `current` should be a separate THREADLOCAL variable;
380 // but it is placed here in order to share cache line with previous fields.
381 ThreadState* current;
382 // This is a slow path flag. On fast path, fast_state.GetIgnoreBit() is read.
383 // We do not distinguish beteween ignoring reads and writes
384 // for better performance.
385 int ignore_reads_and_writes;
386 int ignore_sync;
387 int suppress_reports;
388 // Go does not support ignores.
389#if !SANITIZER_GO
390 IgnoreSet mop_ignore_set;
391 IgnoreSet sync_ignore_set;
392#endif
393 // C/C++ uses fixed size shadow stack embed into Trace.
394 // Go uses malloc-allocated shadow stack with dynamic size.
395 uptr *shadow_stack;
396 uptr *shadow_stack_end;
397 uptr *shadow_stack_pos;
398 u64 *racy_shadow_addr;
399 u64 racy_state[2];
400 MutexSet mset;
401 ThreadClock clock;
402#if !SANITIZER_GO
403 Vector<JmpBuf> jmp_bufs;
404 int ignore_interceptors;
405#endif
406#if TSAN_COLLECT_STATS
407 u64 stat[StatCnt];
408#endif
409 const u32 tid;
410 const int unique_id;
411 bool in_symbolizer;
412 bool in_ignored_lib;
413 bool is_inited;
414 bool is_dead;
415 bool is_freeing;
416 bool is_vptr_access;
417 const uptr stk_addr;
418 const uptr stk_size;
419 const uptr tls_addr;
420 const uptr tls_size;
421 ThreadContext *tctx;
422
423#if SANITIZER_DEBUG && !SANITIZER_GO
424 InternalDeadlockDetector internal_deadlock_detector;
425#endif
426 DDLogicalThread *dd_lt;
427
428 // Current wired Processor, or nullptr. Required to handle any events.
429 Processor *proc1;
430#if !SANITIZER_GO
431 Processor *proc() { return proc1; }
432#else
433 Processor *proc();
434#endif
435
436 atomic_uintptr_t in_signal_handler;
437 ThreadSignalContext *signal_ctx;
438
439#if !SANITIZER_GO
440 u32 last_sleep_stack_id;
441 ThreadClock last_sleep_clock;
442#endif
443
444 // Set in regions of runtime that must be signal-safe and fork-safe.
445 // If set, malloc must not be called.
446 int nomalloc;
447
448 const ReportDesc *current_report;
449
450 explicit ThreadState(Context *ctx, u32 tid, int unique_id, u64 epoch,
451 unsigned reuse_count, uptr stk_addr, uptr stk_size,
452 uptr tls_addr, uptr tls_size);
453};
454
455#if !SANITIZER_GO
456#if SANITIZER_MAC || SANITIZER_ANDROID
457ThreadState *cur_thread();
458void set_cur_thread(ThreadState *thr);
459void cur_thread_finalize();
460inline void cur_thread_init() { }
461#else
462__attribute__((tls_model("initial-exec")))
463extern THREADLOCAL char cur_thread_placeholder[];
464inline ThreadState *cur_thread() {
465 return reinterpret_cast<ThreadState *>(cur_thread_placeholder)->current;
466}
467inline void cur_thread_init() {
468 ThreadState *thr = reinterpret_cast<ThreadState *>(cur_thread_placeholder);
469 if (UNLIKELY(!thr->current))
470 thr->current = thr;
471}
472inline void set_cur_thread(ThreadState *thr) {
473 reinterpret_cast<ThreadState *>(cur_thread_placeholder)->current = thr;
474}
475inline void cur_thread_finalize() { }
476#endif // SANITIZER_MAC || SANITIZER_ANDROID
477#endif // SANITIZER_GO
478
479class ThreadContext final : public ThreadContextBase {
480 public:
481 explicit ThreadContext(int tid);
482 ~ThreadContext();
483 ThreadState *thr;
484 u32 creation_stack_id;
485 SyncClock sync;
486 // Epoch at which the thread had started.
487 // If we see an event from the thread stamped by an older epoch,
488 // the event is from a dead thread that shared tid with this thread.
489 u64 epoch0;
490 u64 epoch1;
491
492 // Override superclass callbacks.
493 void OnDead() override;
494 void OnJoined(void *arg) override;
495 void OnFinished() override;
496 void OnStarted(void *arg) override;
497 void OnCreated(void *arg) override;
498 void OnReset() override;
499 void OnDetached(void *arg) override;
500};
501
502struct RacyStacks {
503 MD5Hash hash[2];
504 bool operator==(const RacyStacks &other) const {
505 if (hash[0] == other.hash[0] && hash[1] == other.hash[1])
506 return true;
507 if (hash[0] == other.hash[1] && hash[1] == other.hash[0])
508 return true;
509 return false;
510 }
511};
512
513struct RacyAddress {
514 uptr addr_min;
515 uptr addr_max;
516};
517
518struct FiredSuppression {
519 ReportType type;
520 uptr pc_or_addr;
521 Suppression *supp;
522};
523
524struct Context {
525 Context();
526
527 bool initialized;
528#if !SANITIZER_GO
529 bool after_multithreaded_fork;
530#endif
531
532 MetaMap metamap;
533
534 Mutex report_mtx;
535 int nreported;
536 int nmissed_expected;
537 atomic_uint64_t last_symbolize_time_ns;
538
539 void *background_thread;
540 atomic_uint32_t stop_background_thread;
541
542 ThreadRegistry *thread_registry;
543
544 Mutex racy_mtx;
545 Vector<RacyStacks> racy_stacks;
546 Vector<RacyAddress> racy_addresses;
547 // Number of fired suppressions may be large enough.
548 Mutex fired_suppressions_mtx;
549 InternalMmapVector<FiredSuppression> fired_suppressions;
550 DDetector *dd;
551
552 ClockAlloc clock_alloc;
553
554 Flags flags;
555
556 u64 stat[StatCnt];
557 u64 int_alloc_cnt[MBlockTypeCount];
558 u64 int_alloc_siz[MBlockTypeCount];
559};
560
561extern Context *ctx; // The one and the only global runtime context.
562
563ALWAYS_INLINE Flags *flags() {
564 return &ctx->flags;
565}
566
567struct ScopedIgnoreInterceptors {
568 ScopedIgnoreInterceptors() {
569#if !SANITIZER_GO
570 cur_thread()->ignore_interceptors++;
571#endif
572 }
573
574 ~ScopedIgnoreInterceptors() {
575#if !SANITIZER_GO
576 cur_thread()->ignore_interceptors--;
577#endif
578 }
579};
580
581const char *GetObjectTypeFromTag(uptr tag);
582const char *GetReportHeaderFromTag(uptr tag);
583uptr TagFromShadowStackFrame(uptr pc);
584
585class ScopedReportBase {
586 public:
587 void AddMemoryAccess(uptr addr, uptr external_tag, Shadow s, StackTrace stack,
588 const MutexSet *mset);
589 void AddStack(StackTrace stack, bool suppressable = false);
590 void AddThread(const ThreadContext *tctx, bool suppressable = false);
591 void AddThread(int unique_tid, bool suppressable = false);
592 void AddUniqueTid(int unique_tid);
593 void AddMutex(const SyncVar *s);
594 u64 AddMutex(u64 id);
595 void AddLocation(uptr addr, uptr size);
596 void AddSleep(u32 stack_id);
597 void SetCount(int count);
598
599 const ReportDesc *GetReport() const;
600
601 protected:
602 ScopedReportBase(ReportType typ, uptr tag);
603 ~ScopedReportBase();
604
605 private:
606 ReportDesc *rep_;
607 // Symbolizer makes lots of intercepted calls. If we try to process them,
608 // at best it will cause deadlocks on internal mutexes.
609 ScopedIgnoreInterceptors ignore_interceptors_;
610
611 void AddDeadMutex(u64 id);
612
613 ScopedReportBase(const ScopedReportBase &) = delete;
614 void operator=(const ScopedReportBase &) = delete;
615};
616
617class ScopedReport : public ScopedReportBase {
618 public:
619 explicit ScopedReport(ReportType typ, uptr tag = kExternalTagNone);
620 ~ScopedReport();
621
622 private:
623 ScopedErrorReportLock lock_;
624};
625
626bool ShouldReport(ThreadState *thr, ReportType typ);
627ThreadContext *IsThreadStackOrTls(uptr addr, bool *is_stack);
628void RestoreStack(int tid, const u64 epoch, VarSizeStackTrace *stk,
629 MutexSet *mset, uptr *tag = nullptr);
630
631// The stack could look like:
632// <start> | <main> | <foo> | tag | <bar>
633// This will extract the tag and keep:
634// <start> | <main> | <foo> | <bar>
635template<typename StackTraceTy>
636void ExtractTagFromStack(StackTraceTy *stack, uptr *tag = nullptr) {
637 if (stack->size < 2) return;
638 uptr possible_tag_pc = stack->trace[stack->size - 2];
639 uptr possible_tag = TagFromShadowStackFrame(possible_tag_pc);
640 if (possible_tag == kExternalTagNone) return;
641 stack->trace_buffer[stack->size - 2] = stack->trace_buffer[stack->size - 1];
642 stack->size -= 1;
643 if (tag) *tag = possible_tag;
644}
645
646template<typename StackTraceTy>
647void ObtainCurrentStack(ThreadState *thr, uptr toppc, StackTraceTy *stack,
648 uptr *tag = nullptr) {
649 uptr size = thr->shadow_stack_pos - thr->shadow_stack;
650 uptr start = 0;
651 if (size + !!toppc > kStackTraceMax) {
652 start = size + !!toppc - kStackTraceMax;
653 size = kStackTraceMax - !!toppc;
654 }
655 stack->Init(&thr->shadow_stack[start], size, toppc);
656 ExtractTagFromStack(stack, tag);
657}
658
659#define GET_STACK_TRACE_FATAL(thr, pc) \
660 VarSizeStackTrace stack; \
661 ObtainCurrentStack(thr, pc, &stack); \
662 stack.ReverseOrder();
663
664#if TSAN_COLLECT_STATS
665void StatAggregate(u64 *dst, u64 *src);
666void StatOutput(u64 *stat);
667#endif
668
669void ALWAYS_INLINE StatInc(ThreadState *thr, StatType typ, u64 n = 1) {
670#if TSAN_COLLECT_STATS
671 thr->stat[typ] += n;
672#endif
673}
674void ALWAYS_INLINE StatSet(ThreadState *thr, StatType typ, u64 n) {
675#if TSAN_COLLECT_STATS
676 thr->stat[typ] = n;
677#endif
678}
679
680void MapShadow(uptr addr, uptr size);
681void MapThreadTrace(uptr addr, uptr size, const char *name);
682void DontNeedShadowFor(uptr addr, uptr size);
683void UnmapShadow(ThreadState *thr, uptr addr, uptr size);
684void InitializeShadowMemory();
685void InitializeInterceptors();
686void InitializeLibIgnore();
687void InitializeDynamicAnnotations();
688
689void ForkBefore(ThreadState *thr, uptr pc);
690void ForkParentAfter(ThreadState *thr, uptr pc);
691void ForkChildAfter(ThreadState *thr, uptr pc);
692
693void ReportRace(ThreadState *thr);
694bool OutputReport(ThreadState *thr, const ScopedReport &srep);
695bool IsFiredSuppression(Context *ctx, ReportType type, StackTrace trace);
696bool IsExpectedReport(uptr addr, uptr size);
697void PrintMatchedBenignRaces();
698
699#if defined(TSAN_DEBUG_OUTPUT) && TSAN_DEBUG_OUTPUT >= 1
700# define DPrintf Printf
701#else
702# define DPrintf(...)
703#endif
704
705#if defined(TSAN_DEBUG_OUTPUT) && TSAN_DEBUG_OUTPUT >= 2
706# define DPrintf2 Printf
707#else
708# define DPrintf2(...)
709#endif
710
711u32 CurrentStackId(ThreadState *thr, uptr pc);
712ReportStack *SymbolizeStackId(u32 stack_id);
713void PrintCurrentStack(ThreadState *thr, uptr pc);
714void PrintCurrentStackSlow(uptr pc); // uses libunwind
715
716void Initialize(ThreadState *thr);
717void MaybeSpawnBackgroundThread();
718int Finalize(ThreadState *thr);
719
720void OnUserAlloc(ThreadState *thr, uptr pc, uptr p, uptr sz, bool write);
721void OnUserFree(ThreadState *thr, uptr pc, uptr p, bool write);
722
723void MemoryAccess(ThreadState *thr, uptr pc, uptr addr,
724 int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic);
725void MemoryAccessImpl(ThreadState *thr, uptr addr,
726 int kAccessSizeLog, bool kAccessIsWrite, bool kIsAtomic,
727 u64 *shadow_mem, Shadow cur);
728void MemoryAccessRange(ThreadState *thr, uptr pc, uptr addr,
729 uptr size, bool is_write);
730void MemoryAccessRangeStep(ThreadState *thr, uptr pc, uptr addr,
731 uptr size, uptr step, bool is_write);
732void UnalignedMemoryAccess(ThreadState *thr, uptr pc, uptr addr,
733 int size, bool kAccessIsWrite, bool kIsAtomic);
734
735const int kSizeLog1 = 0;
736const int kSizeLog2 = 1;
737const int kSizeLog4 = 2;
738const int kSizeLog8 = 3;
739
740void ALWAYS_INLINE MemoryRead(ThreadState *thr, uptr pc,
741 uptr addr, int kAccessSizeLog) {
742 MemoryAccess(thr, pc, addr, kAccessSizeLog, false, false);
743}
744
745void ALWAYS_INLINE MemoryWrite(ThreadState *thr, uptr pc,
746 uptr addr, int kAccessSizeLog) {
747 MemoryAccess(thr, pc, addr, kAccessSizeLog, true, false);
748}
749
750void ALWAYS_INLINE MemoryReadAtomic(ThreadState *thr, uptr pc,
751 uptr addr, int kAccessSizeLog) {
752 MemoryAccess(thr, pc, addr, kAccessSizeLog, false, true);
753}
754
755void ALWAYS_INLINE MemoryWriteAtomic(ThreadState *thr, uptr pc,
756 uptr addr, int kAccessSizeLog) {
757 MemoryAccess(thr, pc, addr, kAccessSizeLog, true, true);
758}
759
760void MemoryResetRange(ThreadState *thr, uptr pc, uptr addr, uptr size);
761void MemoryRangeFreed(ThreadState *thr, uptr pc, uptr addr, uptr size);
762void MemoryRangeImitateWrite(ThreadState *thr, uptr pc, uptr addr, uptr size);
763void MemoryRangeImitateWriteOrResetRange(ThreadState *thr, uptr pc, uptr addr,
764 uptr size);
765
766void ThreadIgnoreBegin(ThreadState *thr, uptr pc, bool save_stack = true);
767void ThreadIgnoreEnd(ThreadState *thr, uptr pc);
768void ThreadIgnoreSyncBegin(ThreadState *thr, uptr pc, bool save_stack = true);
769void ThreadIgnoreSyncEnd(ThreadState *thr, uptr pc);
770
771void FuncEntry(ThreadState *thr, uptr pc);
772void FuncExit(ThreadState *thr);
773
774int ThreadCreate(ThreadState *thr, uptr pc, uptr uid, bool detached);
775void ThreadStart(ThreadState *thr, int tid, tid_t os_id,
776 ThreadType thread_type);
777void ThreadFinish(ThreadState *thr);
778int ThreadConsumeTid(ThreadState *thr, uptr pc, uptr uid);
779void ThreadJoin(ThreadState *thr, uptr pc, int tid);
780void ThreadDetach(ThreadState *thr, uptr pc, int tid);
781void ThreadFinalize(ThreadState *thr);
782void ThreadSetName(ThreadState *thr, const char *name);
783int ThreadCount(ThreadState *thr);
784void ProcessPendingSignals(ThreadState *thr);
785void ThreadNotJoined(ThreadState *thr, uptr pc, int tid, uptr uid);
786
787Processor *ProcCreate();
788void ProcDestroy(Processor *proc);
789void ProcWire(Processor *proc, ThreadState *thr);
790void ProcUnwire(Processor *proc, ThreadState *thr);
791
792// Note: the parameter is called flagz, because flags is already taken
793// by the global function that returns flags.
794void MutexCreate(ThreadState *thr, uptr pc, uptr addr, u32 flagz = 0);
795void MutexDestroy(ThreadState *thr, uptr pc, uptr addr, u32 flagz = 0);
796void MutexPreLock(ThreadState *thr, uptr pc, uptr addr, u32 flagz = 0);
797void MutexPostLock(ThreadState *thr, uptr pc, uptr addr, u32 flagz = 0,
798 int rec = 1);
799int MutexUnlock(ThreadState *thr, uptr pc, uptr addr, u32 flagz = 0);
800void MutexPreReadLock(ThreadState *thr, uptr pc, uptr addr, u32 flagz = 0);
801void MutexPostReadLock(ThreadState *thr, uptr pc, uptr addr, u32 flagz = 0);
802void MutexReadUnlock(ThreadState *thr, uptr pc, uptr addr);
803void MutexReadOrWriteUnlock(ThreadState *thr, uptr pc, uptr addr);
804void MutexRepair(ThreadState *thr, uptr pc, uptr addr); // call on EOWNERDEAD
805void MutexInvalidAccess(ThreadState *thr, uptr pc, uptr addr);
806
807void Acquire(ThreadState *thr, uptr pc, uptr addr);
808// AcquireGlobal synchronizes the current thread with all other threads.
809// In terms of happens-before relation, it draws a HB edge from all threads
810// (where they happen to execute right now) to the current thread. We use it to
811// handle Go finalizers. Namely, finalizer goroutine executes AcquireGlobal
812// right before executing finalizers. This provides a coarse, but simple
813// approximation of the actual required synchronization.
814void AcquireGlobal(ThreadState *thr, uptr pc);
815void Release(ThreadState *thr, uptr pc, uptr addr);
816void ReleaseStoreAcquire(ThreadState *thr, uptr pc, uptr addr);
817void ReleaseStore(ThreadState *thr, uptr pc, uptr addr);
818void AfterSleep(ThreadState *thr, uptr pc);
819void AcquireImpl(ThreadState *thr, uptr pc, SyncClock *c);
820void ReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c);
821void ReleaseStoreAcquireImpl(ThreadState *thr, uptr pc, SyncClock *c);
822void ReleaseStoreImpl(ThreadState *thr, uptr pc, SyncClock *c);
823void AcquireReleaseImpl(ThreadState *thr, uptr pc, SyncClock *c);
824
825// The hacky call uses custom calling convention and an assembly thunk.
826// It is considerably faster that a normal call for the caller
827// if it is not executed (it is intended for slow paths from hot functions).
828// The trick is that the call preserves all registers and the compiler
829// does not treat it as a call.
830// If it does not work for you, use normal call.
831#if !SANITIZER_DEBUG && defined(__x86_64__) && !SANITIZER_MAC
832// The caller may not create the stack frame for itself at all,
833// so we create a reserve stack frame for it (1024b must be enough).
834#define HACKY_CALL(f) \
835 __asm__ __volatile__("sub $1024, %%rsp;" \
836 CFI_INL_ADJUST_CFA_OFFSET(1024) \
837 ".hidden " #f "_thunk;" \
838 "call " #f "_thunk;" \
839 "add $1024, %%rsp;" \
840 CFI_INL_ADJUST_CFA_OFFSET(-1024) \
841 ::: "memory", "cc");
842#else
843#define HACKY_CALL(f) f()
844#endif
845
846void TraceSwitch(ThreadState *thr);
847uptr TraceTopPC(ThreadState *thr);
848uptr TraceSize();
849uptr TraceParts();
850Trace *ThreadTrace(int tid);
851
852extern "C" void __tsan_trace_switch();
853void ALWAYS_INLINE TraceAddEvent(ThreadState *thr, FastState fs,
854 EventType typ, u64 addr) {
855 if (!kCollectHistory)
856 return;
857 DCHECK_GE((int)typ, 0);
858 DCHECK_LE((int)typ, 7);
859 DCHECK_EQ(GetLsb(addr, kEventPCBits), addr);
860 StatInc(thr, StatEvents);
861 u64 pos = fs.GetTracePos();
862 if (UNLIKELY((pos % kTracePartSize) == 0)) {
863#if !SANITIZER_GO
864 HACKY_CALL(__tsan_trace_switch);
865#else
866 TraceSwitch(thr);
867#endif
868 }
869 Event *trace = (Event*)GetThreadTrace(fs.tid());
870 Event *evp = &trace[pos];
871 Event ev = (u64)addr | ((u64)typ << kEventPCBits);
872 *evp = ev;
873}
874
875#if !SANITIZER_GO
876uptr ALWAYS_INLINE HeapEnd() {
877 return HeapMemEnd() + PrimaryAllocator::AdditionalSize();
878}
879#endif
880
881ThreadState *FiberCreate(ThreadState *thr, uptr pc, unsigned flags);
882void FiberDestroy(ThreadState *thr, uptr pc, ThreadState *fiber);
883void FiberSwitch(ThreadState *thr, uptr pc, ThreadState *fiber, unsigned flags);
884
885// These need to match __tsan_switch_to_fiber_* flags defined in
886// tsan_interface.h. See documentation there as well.
887enum FiberSwitchFlags {
888 FiberSwitchFlagNoSync = 1 << 0, // __tsan_switch_to_fiber_no_sync
889};
890
891} // namespace __tsan
892
893#endif // TSAN_RTL_H
894