• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2019 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #ifndef SRC_PROFILING_MEMORY_SHARED_RING_BUFFER_H_
18 #define SRC_PROFILING_MEMORY_SHARED_RING_BUFFER_H_
19 
20 #include "perfetto/ext/base/optional.h"
21 #include "perfetto/ext/base/unix_socket.h"
22 #include "perfetto/ext/base/utils.h"
23 #include "src/profiling/memory/scoped_spinlock.h"
24 #include "src/profiling/memory/util.h"
25 
26 #include <atomic>
27 #include <limits>
28 #include <map>
29 #include <memory>
30 #include <type_traits>
31 
32 #include <stdint.h>
33 
34 namespace perfetto {
35 namespace profiling {
36 
37 // A concurrent, multi-writer single-reader ring buffer FIFO, based on a
38 // circular buffer over shared memory. It has similar semantics to a SEQ_PACKET
39 // + O_NONBLOCK socket, specifically:
40 //
41 // - Writes are atomic, data is either written fully in the buffer or not.
42 // - New writes are discarded if the buffer is full.
43 // - If a write succeeds, the reader is guaranteed to see the whole buffer.
44 // - Reads are atomic, no fragmentation.
45 // - The reader sees writes in write order (% discarding).
46 //
47 // !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
48 // *IMPORTANT*: The ring buffer must be written under the assumption that the
49 // other end modifies arbitrary shared memory without holding the spin-lock.
50 // This means we must make local copies of read and write pointers for doing
51 // bounds checks followed by reads / writes, as they might change in the
52 // meantime.
53 // !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
54 //
55 // TODO:
56 // - Write a benchmark.
57 class SharedRingBuffer {
58  public:
59   class Buffer {
60    public:
Buffer()61     Buffer() {}
Buffer(uint8_t * d,size_t s,uint64_t f)62     Buffer(uint8_t* d, size_t s, uint64_t f)
63         : data(d), size(s), bytes_free(f) {}
64 
65     Buffer(const Buffer&) = delete;
66     Buffer& operator=(const Buffer&) = delete;
67 
68     Buffer(Buffer&&) = default;
69     Buffer& operator=(Buffer&&) = default;
70 
71     explicit operator bool() const { return data != nullptr; }
72 
73     uint8_t* data = nullptr;
74     size_t size = 0;
75     uint64_t bytes_free = 0;
76   };
77 
78   enum ErrorState : uint64_t {
79     kNoError = 0,
80     kHitTimeout = 1,
81     kInvalidStackBounds = 2,
82   };
83 
84   struct Stats {
85     PERFETTO_CROSS_ABI_ALIGNED(uint64_t) bytes_written;
86     PERFETTO_CROSS_ABI_ALIGNED(uint64_t) num_writes_succeeded;
87     PERFETTO_CROSS_ABI_ALIGNED(uint64_t) num_writes_corrupt;
88     PERFETTO_CROSS_ABI_ALIGNED(uint64_t) num_writes_overflow;
89 
90     PERFETTO_CROSS_ABI_ALIGNED(uint64_t) num_reads_succeeded;
91     PERFETTO_CROSS_ABI_ALIGNED(uint64_t) num_reads_corrupt;
92     PERFETTO_CROSS_ABI_ALIGNED(uint64_t) num_reads_nodata;
93 
94     // Fields below get set by GetStats as copies of atomics in MetadataPage.
95     PERFETTO_CROSS_ABI_ALIGNED(uint64_t) failed_spinlocks;
96     PERFETTO_CROSS_ABI_ALIGNED(uint64_t) client_spinlock_blocked_us;
97     PERFETTO_CROSS_ABI_ALIGNED(ErrorState) error_state;
98   };
99 
100   static base::Optional<SharedRingBuffer> Create(size_t);
101   static base::Optional<SharedRingBuffer> Attach(base::ScopedFile);
102 
103   ~SharedRingBuffer();
104   SharedRingBuffer() = default;
105 
106   SharedRingBuffer(SharedRingBuffer&&) noexcept;
107   SharedRingBuffer& operator=(SharedRingBuffer&&) noexcept;
108 
is_valid()109   bool is_valid() const { return !!mem_; }
size()110   size_t size() const { return size_; }
fd()111   int fd() const { return *mem_fd_; }
write_avail()112   size_t write_avail() {
113     auto pos = GetPointerPositions();
114     if (!pos)
115       return 0;
116     return write_avail(*pos);
117   }
118 
119   Buffer BeginWrite(const ScopedSpinlock& spinlock, size_t size);
120   void EndWrite(Buffer buf);
121 
122   Buffer BeginRead();
123   void EndRead(Buffer);
124 
GetStats(ScopedSpinlock & spinlock)125   Stats GetStats(ScopedSpinlock& spinlock) {
126     PERFETTO_DCHECK(spinlock.locked());
127     Stats stats = meta_->stats;
128     stats.failed_spinlocks =
129         meta_->failed_spinlocks.load(std::memory_order_relaxed);
130     stats.error_state = meta_->error_state.load(std::memory_order_relaxed);
131     stats.client_spinlock_blocked_us =
132         meta_->client_spinlock_blocked_us.load(std::memory_order_relaxed);
133     return stats;
134   }
135 
SetErrorState(ErrorState error)136   void SetErrorState(ErrorState error) { meta_->error_state.store(error); }
137 
138   // This is used by the caller to be able to hold the SpinLock after
139   // BeginWrite has returned. This is so that additional bookkeeping can be
140   // done under the lock. This will be used to increment the sequence_number.
AcquireLock(ScopedSpinlock::Mode mode)141   ScopedSpinlock AcquireLock(ScopedSpinlock::Mode mode) {
142     auto lock = ScopedSpinlock(&meta_->spinlock, mode);
143     if (PERFETTO_UNLIKELY(!lock.locked()))
144       meta_->failed_spinlocks.fetch_add(1, std::memory_order_relaxed);
145     return lock;
146   }
147 
AddClientSpinlockBlockedUs(size_t n)148   void AddClientSpinlockBlockedUs(size_t n) {
149     meta_->client_spinlock_blocked_us.fetch_add(n, std::memory_order_relaxed);
150   }
151 
client_spinlock_blocked_us()152   uint64_t client_spinlock_blocked_us() {
153     return meta_->client_spinlock_blocked_us;
154   }
155 
SetShuttingDown()156   void SetShuttingDown() {
157     meta_->shutting_down.store(true, std::memory_order_relaxed);
158   }
159 
shutting_down()160   bool shutting_down() {
161     return meta_->shutting_down.load(std::memory_order_relaxed);
162   }
163 
SetReaderPaused()164   void SetReaderPaused() {
165     meta_->reader_paused.store(true, std::memory_order_relaxed);
166   }
167 
GetAndResetReaderPaused()168   bool GetAndResetReaderPaused() {
169     return meta_->reader_paused.exchange(false, std::memory_order_relaxed);
170   }
171 
InfiniteBufferForTesting()172   void InfiniteBufferForTesting() {
173     // Pretend this buffer is really large, while keeping size_mask_ as
174     // original so it keeps wrapping in circles.
175     size_ = std::numeric_limits<size_t>::max() / 2;
176   }
177 
178   // Exposed for fuzzers.
179   struct MetadataPage {
180     static_assert(std::is_trivially_constructible<Spinlock>::value,
181                   "Spinlock needs to be trivially constructible.");
182     alignas(8) Spinlock spinlock;
183     PERFETTO_CROSS_ABI_ALIGNED(std::atomic<uint64_t>) read_pos;
184     PERFETTO_CROSS_ABI_ALIGNED(std::atomic<uint64_t>) write_pos;
185 
186     PERFETTO_CROSS_ABI_ALIGNED(std::atomic<uint64_t>)
187     client_spinlock_blocked_us;
188     PERFETTO_CROSS_ABI_ALIGNED(std::atomic<uint64_t>) failed_spinlocks;
189     PERFETTO_CROSS_ABI_ALIGNED(std::atomic<ErrorState>) error_state;
190     alignas(sizeof(uint64_t)) std::atomic<bool> shutting_down;
191     alignas(sizeof(uint64_t)) std::atomic<bool> reader_paused;
192     // For stats that are only accessed by a single thread or under the
193     // spinlock, members of this struct are directly modified. Other stats use
194     // the atomics above this struct.
195     //
196     // When the user requests stats, the atomics above get copied into this
197     // struct, which is then returned.
198     alignas(sizeof(uint64_t)) Stats stats;
199   };
200 
201   static_assert(sizeof(MetadataPage) == 144,
202                 "metadata page size needs to be ABI independent");
203 
204  private:
205   struct PointerPositions {
206     uint64_t read_pos;
207     uint64_t write_pos;
208   };
209 
210   struct CreateFlag {};
211   struct AttachFlag {};
212   SharedRingBuffer(const SharedRingBuffer&) = delete;
213   SharedRingBuffer& operator=(const SharedRingBuffer&) = delete;
214   SharedRingBuffer(CreateFlag, size_t size);
SharedRingBuffer(AttachFlag,base::ScopedFile mem_fd)215   SharedRingBuffer(AttachFlag, base::ScopedFile mem_fd) {
216     Initialize(std::move(mem_fd));
217   }
218 
219   void Initialize(base::ScopedFile mem_fd);
220   bool IsCorrupt(const PointerPositions& pos);
221 
GetPointerPositions()222   inline base::Optional<PointerPositions> GetPointerPositions() {
223     PointerPositions pos;
224     // We need to acquire load the write_pos to make sure we observe a
225     // consistent ring buffer in BeginRead, otherwise it is possible that we
226     // observe the write_pos increment, but not the size field write of the
227     // payload.
228     //
229     // This is matched by a release at the end of BeginWrite.
230     pos.write_pos = meta_->write_pos.load(std::memory_order_acquire);
231     pos.read_pos = meta_->read_pos.load(std::memory_order_relaxed);
232 
233     base::Optional<PointerPositions> result;
234     if (IsCorrupt(pos))
235       return result;
236     result = pos;
237     return result;
238   }
239 
set_size(size_t size)240   inline void set_size(size_t size) {
241     size_ = size;
242     size_mask_ = size - 1;
243   }
244 
read_avail(const PointerPositions & pos)245   inline size_t read_avail(const PointerPositions& pos) {
246     PERFETTO_DCHECK(pos.write_pos >= pos.read_pos);
247     auto res = static_cast<size_t>(pos.write_pos - pos.read_pos);
248     PERFETTO_DCHECK(res <= size_);
249     return res;
250   }
251 
write_avail(const PointerPositions & pos)252   inline size_t write_avail(const PointerPositions& pos) {
253     return size_ - read_avail(pos);
254   }
255 
at(uint64_t pos)256   inline uint8_t* at(uint64_t pos) { return mem_ + (pos & size_mask_); }
257 
258   base::ScopedFile mem_fd_;
259   MetadataPage* meta_ = nullptr;  // Start of the mmaped region.
260   uint8_t* mem_ = nullptr;  // Start of the contents (i.e. meta_ + kPageSize).
261 
262   // Size of the ring buffer contents, without including metadata or the 2nd
263   // mmap.
264   size_t size_ = 0;
265   size_t size_mask_ = 0;
266 
267   // Remember to update the move ctor when adding new fields.
268 };
269 
270 }  // namespace profiling
271 }  // namespace perfetto
272 
273 #endif  // SRC_PROFILING_MEMORY_SHARED_RING_BUFFER_H_
274