summaryrefslogblamecommitdiffstats
blob: 55cd6877333a37fb48d0e4687da9496a88b4297e (plain) (tree)
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






































































































































































































































































































































































































































































































































































































































                                                                                               
#include <gtest/gtest.h>
#include <sync/sync.h>
#include <sw_sync.h>
#include <fcntl.h>
#include <vector>
#include <string>
#include <cassert>
#include <iostream>
#include <unistd.h>
#include <thread>
#include <poll.h>
#include <mutex>
#include <algorithm>
#include <tuple>
#include <random>
#include <unordered_map>

// TODO: better stress tests?
// Handle more than 64 fd's simultaneously, i.e. fix sync_fence_info's 4k limit.
// Handle wraparound in timelines like nvidia.

using namespace std;

namespace {

// C++ wrapper class for sync timeline.
class SyncTimeline {
    int m_fd = -1;
    bool m_fdInitialized = false;
public:
    SyncTimeline(const SyncTimeline &) = delete;
    SyncTimeline& operator=(SyncTimeline&) = delete;
    SyncTimeline() noexcept {
        int fd = sw_sync_timeline_create();
        if (fd == -1)
            return;
        m_fdInitialized = true;
        m_fd = fd;
    }
    void destroy() {
        if (m_fdInitialized) {
            close(m_fd);
            m_fd = -1;
            m_fdInitialized = false;
        }
    }
    ~SyncTimeline() {
        destroy();
    }
    bool isValid() const {
        if (m_fdInitialized) {
            int status = fcntl(m_fd, F_GETFD, 0);
            if (status == 0)
                return true;
            else
                return false;
        }
        else {
            return false;
        }
    }
    int getFd() const {
        return m_fd;
    }
    int inc(int val = 1) {
        return sw_sync_timeline_inc(m_fd, val);
    }
};

struct SyncPointInfo {
    std::string driverName;
    std::string objectName;
    uint64_t timeStampNs;
    int status; // 1 sig, 0 active, neg is err
};

// Wrapper class for sync fence.
class SyncFence {
    int m_fd = -1;
    bool m_fdInitialized = false;
    static int s_fenceCount;

    void setFd(int fd) {
        m_fd = fd;
        m_fdInitialized = true;
    }
    void clearFd() {
        m_fd = -1;
        m_fdInitialized = false;
    }
public:
    bool isValid() const {
        if (m_fdInitialized) {
            int status = fcntl(m_fd, F_GETFD, 0);
            if (status == 0)
                return true;
            else
                return false;
        }
        else {
            return false;
        }
    }
    SyncFence& operator=(SyncFence &&rhs) noexcept {
        destroy();
        if (rhs.isValid()) {
            setFd(rhs.getFd());
            rhs.clearFd();
        }
        return *this;
    }
    SyncFence(SyncFence &&fence) noexcept {
        if (fence.isValid()) {
            setFd(fence.getFd());
            fence.clearFd();
        }
    }
    SyncFence(const SyncFence &fence) noexcept {
        // This is ok, as sync fences are immutable after construction, so a dup
        // is basically the same thing as a copy.
        if (fence.isValid()) {
            int fd = dup(fence.getFd());
            if (fd == -1)
                return;
            setFd(fd);
        }
    }
    SyncFence(const SyncTimeline &timeline,
              int value,
              const char *name = nullptr) noexcept {
        std::string autoName = "allocFence";
        autoName += s_fenceCount;
        s_fenceCount++;
        int fd = sw_sync_fence_create(timeline.getFd(), name ? name : autoName.c_str(), value);
        if (fd == -1)
            return;
        setFd(fd);
    }
    SyncFence(const SyncFence &a, const SyncFence &b, const char *name = nullptr) noexcept {
        std::string autoName = "mergeFence";
        autoName += s_fenceCount;
        s_fenceCount++;
        int fd = sync_merge(name ? name : autoName.c_str(), a.getFd(), b.getFd());
        if (fd == -1)
            return;
        setFd(fd);
    }
    SyncFence(const vector<SyncFence> &sources) noexcept {
        assert(sources.size());
        SyncFence temp(*begin(sources));
        for (auto itr = ++begin(sources); itr != end(sources); ++itr) {
            temp = SyncFence(*itr, temp);
        }
        if (temp.isValid()) {
            setFd(temp.getFd());
            temp.clearFd();
        }
    }
    void destroy() {
        if (isValid()) {
            close(m_fd);
            clearFd();
        }
    }
    ~SyncFence() {
        destroy();
    }
    int getFd() const {
        return m_fd;
    }
    int wait(int timeout = -1) {
        return sync_wait(m_fd, timeout);
    }
    vector<SyncPointInfo> getInfo() const {
        struct sync_pt_info *pointInfo = nullptr;
        vector<SyncPointInfo> fenceInfo;
        sync_fence_info_data *info = sync_fence_info(getFd());
        if (!info) {
            return fenceInfo;
        }
        while ((pointInfo = sync_pt_info(info, pointInfo))) {
            fenceInfo.push_back(SyncPointInfo{
                pointInfo->driver_name,
                pointInfo->obj_name,
                pointInfo->timestamp_ns,
                pointInfo->status});
        }
        sync_fence_info_free(info);
        return fenceInfo;
    }
    int getSize() const {
        return getInfo().size();
    }
    int getSignaledCount() const {
        return countWithStatus(1);
    }
    int getActiveCount() const {
        return countWithStatus(0);
    }
    int getErrorCount() const {
        return countWithStatus(-1);
    }
private:
    int countWithStatus(int status) const {
        int count = 0;
        for (auto &info : getInfo()) {
            if (info.status == status) {
                count++;
            }
        }
        return count;
    }
};

int SyncFence::s_fenceCount = 0;

TEST(AllocTest, Timeline) {
    SyncTimeline timeline;
    ASSERT_TRUE(timeline.isValid());
}

TEST(AllocTest, Fence) {
    SyncTimeline timeline;
    ASSERT_TRUE(timeline.isValid());

    SyncFence fence(timeline, 1);
    ASSERT_TRUE(fence.isValid());
}

TEST(AllocTest, FenceNegative) {
    int timeline = sw_sync_timeline_create();
    ASSERT_GT(timeline, 0);

    // bad fd.
    ASSERT_LT(sw_sync_fence_create(-1, "fence", 1), 0);

    // No name - segfaults in user space.
    // Maybe we should be friendlier here?
    /*
    ASSERT_LT(sw_sync_fence_create(timeline, nullptr, 1), 0);
    */
    close(timeline);
}

TEST(FenceTest, OneTimelineWait) {
    SyncTimeline timeline;
    ASSERT_TRUE(timeline.isValid());

    SyncFence fence(timeline, 5);
    ASSERT_TRUE(fence.isValid());

    // Wait on fence until timeout.
    ASSERT_EQ(fence.wait(0), -1);
    ASSERT_EQ(errno, ETIME);

    // Advance timeline from 0 -> 1
    ASSERT_EQ(timeline.inc(1), 0);

    // Wait on fence until timeout.
    ASSERT_EQ(fence.wait(0), -1);
    ASSERT_EQ(errno, ETIME);

    // Signal the fence.
    ASSERT_EQ(timeline.inc(4), 0);

    // Wait successfully.
    ASSERT_EQ(fence.wait(0), 0);

    // Go even futher, and confirm wait still succeeds.
    ASSERT_EQ(timeline.inc(10), 0);
    ASSERT_EQ(fence.wait(0), 0);
}

TEST(FenceTest, OneTimelinePoll) {
    SyncTimeline timeline;
    ASSERT_TRUE(timeline.isValid());

    SyncFence fence(timeline, 100);
    ASSERT_TRUE(fence.isValid());

    fd_set set;
    FD_ZERO(&set);
    FD_SET(fence.getFd(), &set);

    // Poll the fence, and wait till timeout.
    timeval time = {0};
    ASSERT_EQ(select(fence.getFd() + 1, &set, nullptr, nullptr, &time), 0);

    // Advance the timeline.
    timeline.inc(100);
    timeline.inc(100);

    // Select should return that the fd is read for reading.
    FD_ZERO(&set);
    FD_SET(fence.getFd(), &set);

    ASSERT_EQ(select(fence.getFd() + 1, &set, nullptr, nullptr, &time), 1);
    ASSERT_TRUE(FD_ISSET(fence.getFd(), &set));
}

TEST(FenceTest, OneTimelineMerge) {
    SyncTimeline timeline;
    ASSERT_TRUE(timeline.isValid());

    // create fence a,b,c and then merge them all into fence d.
    SyncFence a(timeline, 1), b(timeline, 2), c(timeline, 3);
    ASSERT_TRUE(a.isValid());
    ASSERT_TRUE(b.isValid());
    ASSERT_TRUE(c.isValid());

    SyncFence d({a,b,c});
    ASSERT_TRUE(d.isValid());

    // confirm all fences have one active point (even d).
    ASSERT_EQ(a.getActiveCount(), 1);
    ASSERT_EQ(b.getActiveCount(), 1);
    ASSERT_EQ(c.getActiveCount(), 1);
    ASSERT_EQ(d.getActiveCount(), 1);

    // confirm that d is not signaled until the max of a,b,c
    timeline.inc(1);
    ASSERT_EQ(a.getSignaledCount(), 1);
    ASSERT_EQ(d.getActiveCount(), 1);

    timeline.inc(1);
    ASSERT_EQ(b.getSignaledCount(), 1);
    ASSERT_EQ(d.getActiveCount(), 1);

    timeline.inc(1);
    ASSERT_EQ(c.getSignaledCount(), 1);
    ASSERT_EQ(d.getActiveCount(), 0);
    ASSERT_EQ(d.getSignaledCount(), 1);
}

TEST(FenceTest, MergeSameFence) {
    SyncTimeline timeline;
    ASSERT_TRUE(timeline.isValid());

    SyncFence fence(timeline, 5);
    ASSERT_TRUE(fence.isValid());

    SyncFence selfMergeFence(fence, fence);
    ASSERT_TRUE(selfMergeFence.isValid());

    ASSERT_EQ(selfMergeFence.getSignaledCount(), 0);

    timeline.inc(5);
    ASSERT_EQ(selfMergeFence.getSignaledCount(), 1);
}

TEST(FenceTest, WaitOnDestroyedTimeline) {
    SyncTimeline timeline;
    ASSERT_TRUE(timeline.isValid());

    SyncFence fenceSig(timeline, 100);
    SyncFence fenceKill(timeline, 200);

    // Spawn a thread to wait on a fence when the timeline is killed.
    thread waitThread{
        [&]() {
            ASSERT_EQ(timeline.inc(100), 0);

            ASSERT_EQ(fenceKill.wait(-1), -1);
            ASSERT_EQ(errno, ENOENT);
        }
    };

    // Wait for the thread to spool up.
    fenceSig.wait();

    // Kill the timeline.
    timeline.destroy();

    // wait for the thread to clean up.
    waitThread.join();
}

TEST(FenceTest, PollOnDestroyedTimeline) {
    SyncTimeline timeline;
    ASSERT_TRUE(timeline.isValid());

    SyncFence fenceSig(timeline, 100);
    SyncFence fenceKill(timeline, 200);

    // Spawn a thread to wait on a fence when the timeline is killed.
    thread waitThread{
        [&]() {
            ASSERT_EQ(timeline.inc(100), 0);

            // Wait on the fd.
            struct pollfd fds;
            fds.fd = fenceKill.getFd();
            fds.events = POLLIN | POLLERR;
            ASSERT_EQ(poll(&fds, 1, -1), 1);
            ASSERT_TRUE(fds.revents & POLLERR);
        }
    };

    // Wait for the thread to spool up.
    fenceSig.wait();

    // Kill the timeline.
    timeline.destroy();

    // wait for the thread to clean up.
    waitThread.join();
}

TEST(FenceTest, MultiTimelineWait) {
    SyncTimeline timelineA, timelineB, timelineC;

    SyncFence fenceA(timelineA, 5);
    SyncFence fenceB(timelineB, 5);
    SyncFence fenceC(timelineC, 5);

    // Make a larger fence using 3 other fences from different timelines.
    SyncFence mergedFence({fenceA, fenceB, fenceC});
    ASSERT_TRUE(mergedFence.isValid());

    // Confirm fence isn't signaled
    ASSERT_EQ(mergedFence.getActiveCount(), 3);
    ASSERT_EQ(mergedFence.wait(0), -1);
    ASSERT_EQ(errno, ETIME);

    timelineA.inc(5);
    ASSERT_EQ(mergedFence.getActiveCount(), 2);
    ASSERT_EQ(mergedFence.getSignaledCount(), 1);

    timelineB.inc(5);
    ASSERT_EQ(mergedFence.getActiveCount(), 1);
    ASSERT_EQ(mergedFence.getSignaledCount(), 2);

    timelineC.inc(5);
    ASSERT_EQ(mergedFence.getActiveCount(), 0);
    ASSERT_EQ(mergedFence.getSignaledCount(), 3);

    // confirm you can successfully wait.
    ASSERT_EQ(mergedFence.wait(100), 0);
}

TEST(StressTest, TwoThreadsSharedTimeline) {
    const int iterations = 1 << 16;
    int counter = 0;
    SyncTimeline timeline;
    ASSERT_TRUE(timeline.isValid());

    // Use a single timeline to synchronize two threads
    // hammmering on the same counter.
    auto threadMain = [&](int threadId) {
        for (int i = 0; i < iterations; i++) {
            SyncFence fence(timeline, i * 2 + threadId);
            ASSERT_TRUE(fence.isValid());

            // Wait on the prior thread to complete.
            ASSERT_EQ(fence.wait(), 0);

            // Confirm the previous thread's writes are visible and then inc.
            ASSERT_EQ(counter, i * 2 + threadId);
            counter++;

            // Kick off the other thread.
            ASSERT_EQ(timeline.inc(), 0);
        }
    };

    thread a{threadMain, 0};
    thread b{threadMain, 1};
    a.join();
    b.join();

    // make sure the threads did not trample on one another.
    ASSERT_EQ(counter, iterations * 2);
}

class ConsumerStressTest : public ::testing::TestWithParam<int> {};

TEST_P(ConsumerStressTest, MultiProducerSingleConsumer) {
    mutex lock;
    int counter = 0;
    int iterations = 1 << 12;

    vector<SyncTimeline> producerTimelines(GetParam());
    vector<thread> threads;
    SyncTimeline consumerTimeline;

    // Producer threads run this lambda.
    auto threadMain = [&](int threadId) {
        for (int i = 0; i < iterations; i++) {
            SyncFence fence(consumerTimeline, i);
            ASSERT_TRUE(fence.isValid());

            // Wait for the consumer to finish. Use alternate
            // means of waiting on the fence.
            if ((iterations + threadId) % 8 != 0) {
                ASSERT_EQ(fence.wait(), 0);
            }
            else {
                while (fence.getSignaledCount() != 1) {
                    ASSERT_EQ(fence.getErrorCount(), 0);
                }
            }

            // Every producer increments the counter, the consumer checks + erases it.
            lock.lock();
            counter++;
            lock.unlock();

            ASSERT_EQ(producerTimelines[threadId].inc(), 0);
        }
    };

    for (int i = 0; i < GetParam(); i++) {
        threads.push_back(thread{threadMain, i});
    }

    // Consumer thread runs this loop.
    for (int i = 1; i <= iterations; i++) {
        // Create a fence representing all producers final timelines.
        vector<SyncFence> fences;
        for (auto& timeline : producerTimelines) {
            fences.push_back(SyncFence(timeline, i));
        }
        SyncFence mergeFence(fences);
        ASSERT_TRUE(mergeFence.isValid());

        // Make sure we see an increment from every producer thread. Vary
        // the means by which we wait.
        if (iterations % 8 != 0) {
            ASSERT_EQ(mergeFence.wait(), 0);
        }
        else {
            while (mergeFence.getSignaledCount() != mergeFence.getSize()) {
                ASSERT_EQ(mergeFence.getErrorCount(), 0);
            }
        }
        ASSERT_EQ(counter, GetParam()*i);

        // Release the producer threads.
        ASSERT_EQ(consumerTimeline.inc(), 0);
    }

    for_each(begin(threads), end(threads), [](thread& thread) { thread.join(); });
}
INSTANTIATE_TEST_CASE_P(
    ParameterizedStressTest,
    ConsumerStressTest,
    ::testing::Values(2,4,16));

class MergeStressTest : public ::testing::TestWithParam<tuple<int, int>> {};

template <typename K, typename V> using dict = unordered_map<K,V>;

TEST_P(MergeStressTest, RandomMerge) {
    int timelineCount = get<0>(GetParam());
    int mergeCount = get<1>(GetParam());

    vector<SyncTimeline> timelines(timelineCount);

    default_random_engine generator;
    uniform_int_distribution<int> timelineDist(0, timelines.size()-1);
    uniform_int_distribution<int> syncPointDist(0, numeric_limits<int>::max());

    SyncFence fence(timelines[0], 0);
    ASSERT_TRUE(fence.isValid());

    unordered_map<int, int> fenceMap;
    fenceMap.insert(make_tuple(0, 0));

    // Randomly create syncpoints out of a fixed set of timelines, and merge them together.
    for (int i = 0; i < mergeCount; i++) {

        // Generate syncpoint.
        int timelineOffset = timelineDist(generator);
        const SyncTimeline& timeline = timelines[timelineOffset];
        int syncPoint = syncPointDist(generator);

        // Keep track of the latest syncpoint in each timeline.
        auto itr = fenceMap.find(timelineOffset);
        if (itr == end(fenceMap)) {
            fenceMap.insert(tie(timelineOffset, syncPoint));
        }
        else {
            int oldSyncPoint = itr->second;
            fenceMap.erase(itr);
            fenceMap.insert(tie(timelineOffset, max(syncPoint, oldSyncPoint)));
        }

        // Merge.
        fence = SyncFence(fence, SyncFence(timeline, syncPoint));
        ASSERT_TRUE(fence.isValid());
    }

    // Confirm our map matches the fence.
    ASSERT_EQ(fence.getSize(), fenceMap.size());

    // Trigger the merged fence.
    for (auto& item: fenceMap) {
        ASSERT_EQ(fence.wait(0), -1);
        ASSERT_EQ(errno, ETIME);

        // Increment the timeline to the last syncpoint.
        timelines[item.first].inc(item.second);
    }

    // Check that the fence is triggered.
    ASSERT_EQ(fence.wait(0), 0);
}

INSTANTIATE_TEST_CASE_P(
    ParameterizedMergeStressTest,
    MergeStressTest,
    ::testing::Combine(::testing::Values(16,32), ::testing::Values(32, 1024, 1024*32)));

}