| //===-- sanitizer_deadlock_detector.h ---------------------------*- C++ -*-===// |
| // |
| // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| // See https://llvm.org/LICENSE.txt for license information. |
| // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| // |
| //===----------------------------------------------------------------------===// |
| // |
| // This file is a part of Sanitizer runtime. |
| // The deadlock detector maintains a directed graph of lock acquisitions. |
| // When a lock event happens, the detector checks if the locks already held by |
| // the current thread are reachable from the newly acquired lock. |
| // |
| // The detector can handle only a fixed amount of simultaneously live locks |
| // (a lock is alive if it has been locked at least once and has not been |
| // destroyed). When the maximal number of locks is reached the entire graph |
| // is flushed and the new lock epoch is started. The node ids from the old |
| // epochs can not be used with any of the detector methods except for |
| // nodeBelongsToCurrentEpoch(). |
| // |
| // FIXME: this is work in progress, nothing really works yet. |
| // |
| //===----------------------------------------------------------------------===// |
| |
| #ifndef SANITIZER_DEADLOCK_DETECTOR_H |
| #define SANITIZER_DEADLOCK_DETECTOR_H |
| |
| #include "sanitizer_bvgraph.h" |
| #include "sanitizer_common.h" |
| |
| namespace __sanitizer { |
| |
| // Thread-local state for DeadlockDetector. |
| // It contains the locks currently held by the owning thread. |
| template <class BV> |
| class DeadlockDetectorTLS { |
| public: |
| // No CTOR. |
| void clear() { |
| bv_.clear(); |
| epoch_ = 0; |
| n_recursive_locks = 0; |
| n_all_locks_ = 0; |
| } |
| |
| bool empty() const { return bv_.empty(); } |
| |
| void ensureCurrentEpoch(uptr current_epoch) { |
| if (epoch_ == current_epoch) return; |
| bv_.clear(); |
| epoch_ = current_epoch; |
| n_recursive_locks = 0; |
| n_all_locks_ = 0; |
| } |
| |
| uptr getEpoch() const { return epoch_; } |
| |
| // Returns true if this is the first (non-recursive) acquisition of this lock. |
| bool addLock(uptr lock_id, uptr current_epoch, u32 stk) { |
| CHECK_EQ(epoch_, current_epoch); |
| if (!bv_.setBit(lock_id)) { |
| // The lock is already held by this thread, it must be recursive. |
| CHECK_LT(n_recursive_locks, ARRAY_SIZE(recursive_locks)); |
| recursive_locks[n_recursive_locks++] = lock_id; |
| return false; |
| } |
| CHECK_LT(n_all_locks_, ARRAY_SIZE(all_locks_with_contexts_)); |
| // lock_id < BV::kSize, can cast to a smaller int. |
| u32 lock_id_short = static_cast<u32>(lock_id); |
| LockWithContext l = {lock_id_short, stk}; |
| all_locks_with_contexts_[n_all_locks_++] = l; |
| return true; |
| } |
| |
| void removeLock(uptr lock_id) { |
| if (n_recursive_locks) { |
| for (sptr i = n_recursive_locks - 1; i >= 0; i--) { |
| if (recursive_locks[i] == lock_id) { |
| n_recursive_locks--; |
| Swap(recursive_locks[i], recursive_locks[n_recursive_locks]); |
| return; |
| } |
| } |
| } |
| if (!bv_.clearBit(lock_id)) |
| return; // probably addLock happened before flush |
| if (n_all_locks_) { |
| for (sptr i = n_all_locks_ - 1; i >= 0; i--) { |
| if (all_locks_with_contexts_[i].lock == static_cast<u32>(lock_id)) { |
| Swap(all_locks_with_contexts_[i], |
| all_locks_with_contexts_[n_all_locks_ - 1]); |
| n_all_locks_--; |
| break; |
| } |
| } |
| } |
| } |
| |
| u32 findLockContext(uptr lock_id) { |
| for (uptr i = 0; i < n_all_locks_; i++) |
| if (all_locks_with_contexts_[i].lock == static_cast<u32>(lock_id)) |
| return all_locks_with_contexts_[i].stk; |
| return 0; |
| } |
| |
| const BV &getLocks(uptr current_epoch) const { |
| CHECK_EQ(epoch_, current_epoch); |
| return bv_; |
| } |
| |
| uptr getNumLocks() const { return n_all_locks_; } |
| uptr getLock(uptr idx) const { return all_locks_with_contexts_[idx].lock; } |
| |
| private: |
| BV bv_; |
| uptr epoch_; |
| uptr recursive_locks[64]; |
| uptr n_recursive_locks; |
| struct LockWithContext { |
| u32 lock; |
| u32 stk; |
| }; |
| LockWithContext all_locks_with_contexts_[64]; |
| uptr n_all_locks_; |
| }; |
| |
| // DeadlockDetector. |
| // For deadlock detection to work we need one global DeadlockDetector object |
| // and one DeadlockDetectorTLS object per evey thread. |
| // This class is not thread safe, all concurrent accesses should be guarded |
| // by an external lock. |
| // Most of the methods of this class are not thread-safe (i.e. should |
| // be protected by an external lock) unless explicitly told otherwise. |
| template <class BV> |
| class DeadlockDetector { |
| public: |
| typedef BV BitVector; |
| |
| uptr size() const { return g_.size(); } |
| |
| // No CTOR. |
| void clear() { |
| current_epoch_ = 0; |
| available_nodes_.clear(); |
| recycled_nodes_.clear(); |
| g_.clear(); |
| n_edges_ = 0; |
| } |
| |
| // Allocate new deadlock detector node. |
| // If we are out of available nodes first try to recycle some. |
| // If there is nothing to recycle, flush the graph and increment the epoch. |
| // Associate 'data' (opaque user's object) with the new node. |
| uptr newNode(uptr data) { |
| if (!available_nodes_.empty()) |
| return getAvailableNode(data); |
| if (!recycled_nodes_.empty()) { |
| for (sptr i = n_edges_ - 1; i >= 0; i--) { |
| if (recycled_nodes_.getBit(edges_[i].from) || |
| recycled_nodes_.getBit(edges_[i].to)) { |
| Swap(edges_[i], edges_[n_edges_ - 1]); |
| n_edges_--; |
| } |
| } |
| CHECK(available_nodes_.empty()); |
| // removeEdgesFrom was called in removeNode. |
| g_.removeEdgesTo(recycled_nodes_); |
| available_nodes_.setUnion(recycled_nodes_); |
| recycled_nodes_.clear(); |
| return getAvailableNode(data); |
| } |
| // We are out of vacant nodes. Flush and increment the current_epoch_. |
| current_epoch_ += size(); |
| recycled_nodes_.clear(); |
| available_nodes_.setAll(); |
| g_.clear(); |
| n_edges_ = 0; |
| return getAvailableNode(data); |
| } |
| |
| // Get data associated with the node created by newNode(). |
| uptr getData(uptr node) const { return data_[nodeToIndex(node)]; } |
| |
| bool nodeBelongsToCurrentEpoch(uptr node) { |
| return node && (node / size() * size()) == current_epoch_; |
| } |
| |
| void removeNode(uptr node) { |
| uptr idx = nodeToIndex(node); |
| CHECK(!available_nodes_.getBit(idx)); |
| CHECK(recycled_nodes_.setBit(idx)); |
| g_.removeEdgesFrom(idx); |
| } |
| |
| void ensureCurrentEpoch(DeadlockDetectorTLS<BV> *dtls) { |
| dtls->ensureCurrentEpoch(current_epoch_); |
| } |
| |
| // Returns true if there is a cycle in the graph after this lock event. |
| // Ideally should be called before the lock is acquired so that we can |
| // report a deadlock before a real deadlock happens. |
| bool onLockBefore(DeadlockDetectorTLS<BV> *dtls, uptr cur_node) { |
| ensureCurrentEpoch(dtls); |
| uptr cur_idx = nodeToIndex(cur_node); |
| return g_.isReachable(cur_idx, dtls->getLocks(current_epoch_)); |
| } |
| |
| u32 findLockContext(DeadlockDetectorTLS<BV> *dtls, uptr node) { |
| return dtls->findLockContext(nodeToIndex(node)); |
| } |
| |
| // Add cur_node to the set of locks held currently by dtls. |
| void onLockAfter(DeadlockDetectorTLS<BV> *dtls, uptr cur_node, u32 stk = 0) { |
| ensureCurrentEpoch(dtls); |
| uptr cur_idx = nodeToIndex(cur_node); |
| dtls->addLock(cur_idx, current_epoch_, stk); |
| } |
| |
| // Experimental *racy* fast path function. |
| // Returns true if all edges from the currently held locks to cur_node exist. |
| bool hasAllEdges(DeadlockDetectorTLS<BV> *dtls, uptr cur_node) { |
| uptr local_epoch = dtls->getEpoch(); |
| // Read from current_epoch_ is racy. |
| if (cur_node && local_epoch == current_epoch_ && |
| local_epoch == nodeToEpoch(cur_node)) { |
| uptr cur_idx = nodeToIndexUnchecked(cur_node); |
| for (uptr i = 0, n = dtls->getNumLocks(); i < n; i++) { |
| if (!g_.hasEdge(dtls->getLock(i), cur_idx)) |
| return false; |
| } |
| return true; |
| } |
| return false; |
| } |
| |
| // Adds edges from currently held locks to cur_node, |
| // returns the number of added edges, and puts the sources of added edges |
| // into added_edges[]. |
| // Should be called before onLockAfter. |
| uptr addEdges(DeadlockDetectorTLS<BV> *dtls, uptr cur_node, u32 stk, |
| int unique_tid) { |
| ensureCurrentEpoch(dtls); |
| uptr cur_idx = nodeToIndex(cur_node); |
| uptr added_edges[40]; |
| uptr n_added_edges = g_.addEdges(dtls->getLocks(current_epoch_), cur_idx, |
| added_edges, ARRAY_SIZE(added_edges)); |
| for (uptr i = 0; i < n_added_edges; i++) { |
| if (n_edges_ < ARRAY_SIZE(edges_)) { |
| Edge e = {(u16)added_edges[i], (u16)cur_idx, |
| dtls->findLockContext(added_edges[i]), stk, |
| unique_tid}; |
| edges_[n_edges_++] = e; |
| } |
| } |
| return n_added_edges; |
| } |
| |
| bool findEdge(uptr from_node, uptr to_node, u32 *stk_from, u32 *stk_to, |
| int *unique_tid) { |
| uptr from_idx = nodeToIndex(from_node); |
| uptr to_idx = nodeToIndex(to_node); |
| for (uptr i = 0; i < n_edges_; i++) { |
| if (edges_[i].from == from_idx && edges_[i].to == to_idx) { |
| *stk_from = edges_[i].stk_from; |
| *stk_to = edges_[i].stk_to; |
| *unique_tid = edges_[i].unique_tid; |
| return true; |
| } |
| } |
| return false; |
| } |
| |
| // Test-only function. Handles the before/after lock events, |
| // returns true if there is a cycle. |
| bool onLock(DeadlockDetectorTLS<BV> *dtls, uptr cur_node, u32 stk = 0) { |
| ensureCurrentEpoch(dtls); |
| bool is_reachable = !isHeld(dtls, cur_node) && onLockBefore(dtls, cur_node); |
| addEdges(dtls, cur_node, stk, 0); |
| onLockAfter(dtls, cur_node, stk); |
| return is_reachable; |
| } |
| |
| // Handles the try_lock event, returns false. |
| // When a try_lock event happens (i.e. a try_lock call succeeds) we need |
| // to add this lock to the currently held locks, but we should not try to |
| // change the lock graph or to detect a cycle. We may want to investigate |
| // whether a more aggressive strategy is possible for try_lock. |
| bool onTryLock(DeadlockDetectorTLS<BV> *dtls, uptr cur_node, u32 stk = 0) { |
| ensureCurrentEpoch(dtls); |
| uptr cur_idx = nodeToIndex(cur_node); |
| dtls->addLock(cur_idx, current_epoch_, stk); |
| return false; |
| } |
| |
| // Returns true iff dtls is empty (no locks are currently held) and we can |
| // add the node to the currently held locks w/o changing the global state. |
| // This operation is thread-safe as it only touches the dtls. |
| bool onFirstLock(DeadlockDetectorTLS<BV> *dtls, uptr node, u32 stk = 0) { |
| if (!dtls->empty()) return false; |
| if (dtls->getEpoch() && dtls->getEpoch() == nodeToEpoch(node)) { |
| dtls->addLock(nodeToIndexUnchecked(node), nodeToEpoch(node), stk); |
| return true; |
| } |
| return false; |
| } |
| |
| // Finds a path between the lock 'cur_node' (currently not held in dtls) |
| // and some currently held lock, returns the length of the path |
| // or 0 on failure. |
| uptr findPathToLock(DeadlockDetectorTLS<BV> *dtls, uptr cur_node, uptr *path, |
| uptr path_size) { |
| tmp_bv_.copyFrom(dtls->getLocks(current_epoch_)); |
| uptr idx = nodeToIndex(cur_node); |
| CHECK(!tmp_bv_.getBit(idx)); |
| uptr res = g_.findShortestPath(idx, tmp_bv_, path, path_size); |
| for (uptr i = 0; i < res; i++) |
| path[i] = indexToNode(path[i]); |
| if (res) |
| CHECK_EQ(path[0], cur_node); |
| return res; |
| } |
| |
| // Handle the unlock event. |
| // This operation is thread-safe as it only touches the dtls. |
| void onUnlock(DeadlockDetectorTLS<BV> *dtls, uptr node) { |
| if (dtls->getEpoch() == nodeToEpoch(node)) |
| dtls->removeLock(nodeToIndexUnchecked(node)); |
| } |
| |
| // Tries to handle the lock event w/o writing to global state. |
| // Returns true on success. |
| // This operation is thread-safe as it only touches the dtls |
| // (modulo racy nature of hasAllEdges). |
| bool onLockFast(DeadlockDetectorTLS<BV> *dtls, uptr node, u32 stk = 0) { |
| if (hasAllEdges(dtls, node)) { |
| dtls->addLock(nodeToIndexUnchecked(node), nodeToEpoch(node), stk); |
| return true; |
| } |
| return false; |
| } |
| |
| bool isHeld(DeadlockDetectorTLS<BV> *dtls, uptr node) const { |
| return dtls->getLocks(current_epoch_).getBit(nodeToIndex(node)); |
| } |
| |
| uptr testOnlyGetEpoch() const { return current_epoch_; } |
| bool testOnlyHasEdge(uptr l1, uptr l2) { |
| return g_.hasEdge(nodeToIndex(l1), nodeToIndex(l2)); |
| } |
| // idx1 and idx2 are raw indices to g_, not lock IDs. |
| bool testOnlyHasEdgeRaw(uptr idx1, uptr idx2) { |
| return g_.hasEdge(idx1, idx2); |
| } |
| |
| void Print() { |
| for (uptr from = 0; from < size(); from++) |
| for (uptr to = 0; to < size(); to++) |
| if (g_.hasEdge(from, to)) |
| Printf(" %zx => %zx\n", from, to); |
| } |
| |
| private: |
| void check_idx(uptr idx) const { CHECK_LT(idx, size()); } |
| |
| void check_node(uptr node) const { |
| CHECK_GE(node, size()); |
| CHECK_EQ(current_epoch_, nodeToEpoch(node)); |
| } |
| |
| uptr indexToNode(uptr idx) const { |
| check_idx(idx); |
| return idx + current_epoch_; |
| } |
| |
| uptr nodeToIndexUnchecked(uptr node) const { return node % size(); } |
| |
| uptr nodeToIndex(uptr node) const { |
| check_node(node); |
| return nodeToIndexUnchecked(node); |
| } |
| |
| uptr nodeToEpoch(uptr node) const { return node / size() * size(); } |
| |
| uptr getAvailableNode(uptr data) { |
| uptr idx = available_nodes_.getAndClearFirstOne(); |
| data_[idx] = data; |
| return indexToNode(idx); |
| } |
| |
| struct Edge { |
| u16 from; |
| u16 to; |
| u32 stk_from; |
| u32 stk_to; |
| int unique_tid; |
| }; |
| |
| uptr current_epoch_; |
| BV available_nodes_; |
| BV recycled_nodes_; |
| BV tmp_bv_; |
| BVGraph<BV> g_; |
| uptr data_[BV::kSize]; |
| Edge edges_[BV::kSize * 32]; |
| uptr n_edges_; |
| }; |
| |
| } // namespace __sanitizer |
| |
| #endif // SANITIZER_DEADLOCK_DETECTOR_H |