Index/src / corosio / src / detail / select / scheduler.cpp
Prev Index Next

src/corosio/src/detail/select/scheduler.cpp

73.4% Lines (268/365) 87.9% Functions (29/33) 62.2% Branches (166/267)
src/corosio/src/detail/select/scheduler.cpp
Line Branch Hits Source Code
1 //
2 // Copyright (c) 2026 Steve Gerbino
3 //
4 // Distributed under the Boost Software License, Version 1.0. (See accompanying
5 // file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
6 //
7 // Official repository: https://github.com/cppalliance/corosio
8 //
9
10 #include <boost/corosio/detail/platform.hpp>
11
12 #if BOOST_COROSIO_HAS_SELECT
13
14 #include "src/detail/select/scheduler.hpp"
15 #include "src/detail/select/op.hpp"
16 #include "src/detail/timer_service.hpp"
17 #include "src/detail/make_err.hpp"
18 #include "src/detail/posix/resolver_service.hpp"
19 #include "src/detail/posix/signals.hpp"
20
21 #include <boost/corosio/detail/except.hpp>
22 #include <boost/corosio/detail/thread_local_ptr.hpp>
23
24 #include <chrono>
25 #include <limits>
26
27 #include <errno.h>
28 #include <fcntl.h>
29 #include <sys/select.h>
30 #include <sys/socket.h>
31 #include <unistd.h>
32
33 /*
34 select Scheduler - Single Reactor Model
35 =======================================
36
37 This scheduler mirrors the epoll_scheduler design but uses select() instead
38 of epoll for I/O multiplexing. The thread coordination strategy is identical:
39 one thread becomes the "reactor" while others wait on a condition variable.
40
41 Thread Model
42 ------------
43 - ONE thread runs select() at a time (the reactor thread)
44 - OTHER threads wait on wakeup_event_ (condition variable) for handlers
45 - When work is posted, exactly one waiting thread wakes via notify_one()
46
47 Key Differences from epoll
48 --------------------------
49 - Uses self-pipe instead of eventfd for interruption (more portable)
50 - fd_set rebuilding each iteration (O(n) vs O(1) for epoll)
51 - FD_SETSIZE limit (~1024 fds on most systems)
52 - Level-triggered only (no edge-triggered mode)
53
54 Self-Pipe Pattern
55 -----------------
56 To interrupt a blocking select() call (e.g., when work is posted or a timer
57 expires), we write a byte to pipe_fds_[1]. The read end pipe_fds_[0] is
58 always in the read_fds set, so select() returns immediately. We drain the
59 pipe to clear the readable state.
60
61 fd-to-op Mapping
62 ----------------
63 We use an unordered_map<int, fd_state> to track which operations are
64 registered for each fd. This allows O(1) lookup when select() returns
65 ready fds. Each fd can have at most one read op and one write op registered.
66 */
67
68 namespace boost::corosio::detail {
69
70 namespace {
71
72 struct scheduler_context
73 {
74 select_scheduler const* key;
75 scheduler_context* next;
76 };
77
78 corosio::detail::thread_local_ptr<scheduler_context> context_stack;
79
80 struct thread_context_guard
81 {
82 scheduler_context frame_;
83
84 120 explicit thread_context_guard(
85 select_scheduler const* ctx) noexcept
86 120 : frame_{ctx, context_stack.get()}
87 {
88 120 context_stack.set(&frame_);
89 120 }
90
91 120 ~thread_context_guard() noexcept
92 {
93 120 context_stack.set(frame_.next);
94 120 }
95 };
96
97 } // namespace
98
99 133 select_scheduler::
100 select_scheduler(
101 capy::execution_context& ctx,
102 133 int)
103 133 : pipe_fds_{-1, -1}
104 133 , outstanding_work_(0)
105 133 , stopped_(false)
106 133 , shutdown_(false)
107 133 , max_fd_(-1)
108 133 , reactor_running_(false)
109 133 , reactor_interrupted_(false)
110 266 , idle_thread_count_(0)
111 {
112 // Create self-pipe for interrupting select()
113
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 133 times.
 133 if (::pipe(pipe_fds_) < 0)
114 ✗ detail::throw_system_error(make_err(errno), "pipe");
115
116 // Set both ends to non-blocking and close-on-exec
117
2/2
✓ Branch 0 taken 266 times.
✓ Branch 1 taken 133 times.
 399 for (int i = 0; i < 2; ++i)
118 {
119
1/1
✓ Branch 1 taken 266 times.
 266 int flags = ::fcntl(pipe_fds_[i], F_GETFL, 0);
120
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 266 times.
 266 if (flags == -1)
121 {
122 ✗ int errn = errno;
123 ✗ ::close(pipe_fds_[0]);
124 ✗ ::close(pipe_fds_[1]);
125 ✗ detail::throw_system_error(make_err(errn), "fcntl F_GETFL");
126 }
127
2/3
✓ Branch 1 taken 266 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 266 times.
 266 if (::fcntl(pipe_fds_[i], F_SETFL, flags | O_NONBLOCK) == -1)
128 {
129 ✗ int errn = errno;
130 ✗ ::close(pipe_fds_[0]);
131 ✗ ::close(pipe_fds_[1]);
132 ✗ detail::throw_system_error(make_err(errn), "fcntl F_SETFL");
133 }
134
2/3
✓ Branch 1 taken 266 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 266 times.
 266 if (::fcntl(pipe_fds_[i], F_SETFD, FD_CLOEXEC) == -1)
135 {
136 ✗ int errn = errno;
137 ✗ ::close(pipe_fds_[0]);
138 ✗ ::close(pipe_fds_[1]);
139 ✗ detail::throw_system_error(make_err(errn), "fcntl F_SETFD");
140 }
141 }
142
143
1/1
✓ Branch 1 taken 133 times.
 133 timer_svc_ = &get_timer_service(ctx, *this);
144
1/1
✓ Branch 3 taken 133 times.
 133 timer_svc_->set_on_earliest_changed(
145 timer_service::callback(
146 this,
147 3773 [](void* p) { static_cast<select_scheduler*>(p)->interrupt_reactor(); }));
148
149 // Initialize resolver service
150
1/1
✓ Branch 1 taken 133 times.
 133 get_resolver_service(ctx, *this);
151
152 // Initialize signal service
153
1/1
✓ Branch 1 taken 133 times.
 133 get_signal_service(ctx, *this);
154
155 // Push task sentinel to interleave reactor runs with handler execution
156 133 completed_ops_.push(&task_op_);
157 133 }
158
159 266 select_scheduler::
160 133 ~select_scheduler()
161 {
162
1/2
✓ Branch 0 taken 133 times.
✗ Branch 1 not taken.
 133 if (pipe_fds_[0] >= 0)
163 133 ::close(pipe_fds_[0]);
164
1/2
✓ Branch 0 taken 133 times.
✗ Branch 1 not taken.
 133 if (pipe_fds_[1] >= 0)
165 133 ::close(pipe_fds_[1]);
166 266 }
167
168 void
169 133 select_scheduler::
170 shutdown()
171 {
172 {
173
1/1
✓ Branch 1 taken 133 times.
 133 std::unique_lock lock(mutex_);
174 133 shutdown_ = true;
175
176
2/2
✓ Branch 1 taken 133 times.
✓ Branch 2 taken 133 times.
 266 while (auto* h = completed_ops_.pop())
177 {
178
1/2
✓ Branch 0 taken 133 times.
✗ Branch 1 not taken.
 133 if (h == &task_op_)
179 133 continue;
180 ✗ lock.unlock();
181 ✗ h->destroy();
182 ✗ lock.lock();
183 133 }
184 133 }
185
186 133 outstanding_work_.store(0, std::memory_order_release);
187
188
1/2
✓ Branch 0 taken 133 times.
✗ Branch 1 not taken.
 133 if (pipe_fds_[1] >= 0)
189 133 interrupt_reactor();
190
191 133 wakeup_event_.notify_all();
192 133 }
193
194 void
195 4112 select_scheduler::
196 post(std::coroutine_handle<> h) const
197 {
198 struct post_handler final
199 : scheduler_op
200 {
201 std::coroutine_handle<> h_;
202
203 explicit
204 4112 post_handler(std::coroutine_handle<> h)
205 4112 : h_(h)
206 {
207 4112 }
208
209 8224 ~post_handler() = default;
210
211 4112 void operator()() override
212 {
213 4112 auto h = h_;
214
1/2
✓ Branch 0 taken 4112 times.
✗ Branch 1 not taken.
 4112 delete this;
215
1/1
✓ Branch 1 taken 4112 times.
 4112 h.resume();
216 4112 }
217
218 ✗ void destroy() override
219 {
220 ✗ delete this;
221 ✗ }
222 };
223
224
1/1
✓ Branch 1 taken 4112 times.
 4112 auto ph = std::make_unique<post_handler>(h);
225 4112 outstanding_work_.fetch_add(1, std::memory_order_relaxed);
226
227
1/1
✓ Branch 1 taken 4112 times.
 4112 std::unique_lock lock(mutex_);
228 4112 completed_ops_.push(ph.release());
229
1/1
✓ Branch 1 taken 4112 times.
 4112 wake_one_thread_and_unlock(lock);
230 4112 }
231
232 void
233 247068 select_scheduler::
234 post(scheduler_op* h) const
235 {
236 247068 outstanding_work_.fetch_add(1, std::memory_order_relaxed);
237
238
1/1
✓ Branch 1 taken 247068 times.
 247068 std::unique_lock lock(mutex_);
239 247068 completed_ops_.push(h);
240
1/1
✓ Branch 1 taken 247068 times.
 247068 wake_one_thread_and_unlock(lock);
241 247068 }
242
243 void
244 4351 select_scheduler::
245 on_work_started() noexcept
246 {
247 4351 outstanding_work_.fetch_add(1, std::memory_order_relaxed);
248 4351 }
249
250 void
251 4345 select_scheduler::
252 on_work_finished() noexcept
253 {
254
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4345 times.
 8690 if (outstanding_work_.fetch_sub(1, std::memory_order_acq_rel) == 1)
255 ✗ stop();
256 4345 }
257
258 bool
259 556 select_scheduler::
260 running_in_this_thread() const noexcept
261 {
262
2/2
✓ Branch 1 taken 372 times.
✓ Branch 2 taken 184 times.
 556 for (auto* c = context_stack.get(); c != nullptr; c = c->next)
263
1/2
✓ Branch 0 taken 372 times.
✗ Branch 1 not taken.
 372 if (c->key == this)
264 372 return true;
265 184 return false;
266 }
267
268 void
269 3 select_scheduler::
270 stop()
271 {
272 3 bool expected = false;
273
1/2
✓ Branch 1 taken 3 times.
✗ Branch 2 not taken.
 3 if (stopped_.compare_exchange_strong(expected, true,
274 std::memory_order_release, std::memory_order_relaxed))
275 {
276 // Wake all threads so they notice stopped_ and exit
277 {
278
1/1
✓ Branch 1 taken 3 times.
 3 std::lock_guard lock(mutex_);
279 3 wakeup_event_.notify_all();
280 3 }
281
1/1
✓ Branch 1 taken 3 times.
 3 interrupt_reactor();
282 }
283 3 }
284
285 bool
286 3 select_scheduler::
287 stopped() const noexcept
288 {
289 3 return stopped_.load(std::memory_order_acquire);
290 }
291
292 void
293 34 select_scheduler::
294 restart()
295 {
296 34 stopped_.store(false, std::memory_order_release);
297 34 }
298
299 std::size_t
300 96 select_scheduler::
301 run()
302 {
303
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 96 times.
 96 if (stopped_.load(std::memory_order_acquire))
304 ✗ return 0;
305
306
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 96 times.
 192 if (outstanding_work_.load(std::memory_order_acquire) == 0)
307 {
308 ✗ stop();
309 ✗ return 0;
310 }
311
312 96 thread_context_guard ctx(this);
313
314 96 std::size_t n = 0;
315
3/3
✓ Branch 1 taken 258482 times.
✓ Branch 3 taken 258386 times.
✓ Branch 4 taken 96 times.
 258482 while (do_one(-1))
316
1/2
✓ Branch 1 taken 258386 times.
✗ Branch 2 not taken.
 258386 if (n != (std::numeric_limits<std::size_t>::max)())
317 258386 ++n;
318 96 return n;
319 96 }
320
321 std::size_t
322 ✗ select_scheduler::
323 run_one()
324 {
325 ✗ if (stopped_.load(std::memory_order_acquire))
326 ✗ return 0;
327
328 ✗ if (outstanding_work_.load(std::memory_order_acquire) == 0)
329 {
330 ✗ stop();
331 ✗ return 0;
332 }
333
334 ✗ thread_context_guard ctx(this);
335 ✗ return do_one(-1);
336 ✗ }
337
338 std::size_t
339 27 select_scheduler::
340 wait_one(long usec)
341 {
342
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 27 times.
 27 if (stopped_.load(std::memory_order_acquire))
343 ✗ return 0;
344
345
2/2
✓ Branch 1 taken 3 times.
✓ Branch 2 taken 24 times.
 54 if (outstanding_work_.load(std::memory_order_acquire) == 0)
346 {
347
1/1
✓ Branch 1 taken 3 times.
 3 stop();
348 3 return 0;
349 }
350
351 24 thread_context_guard ctx(this);
352
1/1
✓ Branch 1 taken 24 times.
 24 return do_one(usec);
353 24 }
354
355 std::size_t
356 ✗ select_scheduler::
357 poll()
358 {
359 ✗ if (stopped_.load(std::memory_order_acquire))
360 ✗ return 0;
361
362 ✗ if (outstanding_work_.load(std::memory_order_acquire) == 0)
363 {
364 ✗ stop();
365 ✗ return 0;
366 }
367
368 ✗ thread_context_guard ctx(this);
369
370 ✗ std::size_t n = 0;
371 ✗ while (do_one(0))
372 ✗ if (n != (std::numeric_limits<std::size_t>::max)())
373 ✗ ++n;
374 ✗ return n;
375 ✗ }
376
377 std::size_t
378 ✗ select_scheduler::
379 poll_one()
380 {
381 ✗ if (stopped_.load(std::memory_order_acquire))
382 ✗ return 0;
383
384 ✗ if (outstanding_work_.load(std::memory_order_acquire) == 0)
385 {
386 ✗ stop();
387 ✗ return 0;
388 }
389
390 ✗ thread_context_guard ctx(this);
391 ✗ return do_one(0);
392 ✗ }
393
394 void
395 7394 select_scheduler::
396 register_fd(int fd, select_op* op, int events) const
397 {
398 // Validate fd is within select() limits
399
2/4
✓ Branch 0 taken 7394 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 7394 times.
 7394 if (fd < 0 || fd >= FD_SETSIZE)
400 ✗ detail::throw_system_error(make_err(EINVAL), "select: fd out of range");
401
402 {
403
1/1
✓ Branch 1 taken 7394 times.
 7394 std::lock_guard lock(mutex_);
404
405
1/1
✓ Branch 1 taken 7394 times.
 7394 auto& state = registered_fds_[fd];
406
2/2
✓ Branch 0 taken 3839 times.
✓ Branch 1 taken 3555 times.
 7394 if (events & event_read)
407 3839 state.read_op = op;
408
2/2
✓ Branch 0 taken 3555 times.
✓ Branch 1 taken 3839 times.
 7394 if (events & event_write)
409 3555 state.write_op = op;
410
411
2/2
✓ Branch 0 taken 229 times.
✓ Branch 1 taken 7165 times.
 7394 if (fd > max_fd_)
412 229 max_fd_ = fd;
413 7394 }
414
415 // Wake the reactor so a thread blocked in select() rebuilds its fd_sets
416 // with the newly registered fd.
417 7394 interrupt_reactor();
418 7394 }
419
420 void
421 7325 select_scheduler::
422 deregister_fd(int fd, int events) const
423 {
424
1/1
✓ Branch 1 taken 7325 times.
 7325 std::lock_guard lock(mutex_);
425
426
1/1
✓ Branch 1 taken 7325 times.
 7325 auto it = registered_fds_.find(fd);
427
2/2
✓ Branch 2 taken 7161 times.
✓ Branch 3 taken 164 times.
 7325 if (it == registered_fds_.end())
428 7161 return;
429
430
1/2
✓ Branch 0 taken 164 times.
✗ Branch 1 not taken.
 164 if (events & event_read)
431 164 it->second.read_op = nullptr;
432
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 164 times.
 164 if (events & event_write)
433 ✗ it->second.write_op = nullptr;
434
435 // Remove entry if both are null
436
3/6
✓ Branch 1 taken 164 times.
✗ Branch 2 not taken.
✓ Branch 4 taken 164 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 164 times.
✗ Branch 7 not taken.
 164 if (!it->second.read_op && !it->second.write_op)
437 {
438
1/1
✓ Branch 1 taken 164 times.
 164 registered_fds_.erase(it);
439
440 // Recalculate max_fd_ if needed
441
2/2
✓ Branch 0 taken 163 times.
✓ Branch 1 taken 1 time.
 164 if (fd == max_fd_)
442 {
443 163 max_fd_ = pipe_fds_[0]; // At minimum, the pipe read end
444
1/2
✗ Branch 7 not taken.
✓ Branch 8 taken 163 times.
 163 for (auto& [registered_fd, state] : registered_fds_)
445 {
446 ✗ if (registered_fd > max_fd_)
447 ✗ max_fd_ = registered_fd;
448 }
449 }
450 }
451 7325 }
452
453 void
454 7394 select_scheduler::
455 work_started() const noexcept
456 {
457 7394 outstanding_work_.fetch_add(1, std::memory_order_relaxed);
458 7394 }
459
460 void
461 258580 select_scheduler::
462 work_finished() const noexcept
463 {
464
2/2
✓ Branch 0 taken 99 times.
✓ Branch 1 taken 258481 times.
 517160 if (outstanding_work_.fetch_sub(1, std::memory_order_acq_rel) == 1)
465 {
466 // Last work item completed - wake all threads so they can exit.
467 99 std::unique_lock lock(mutex_);
468 99 wakeup_event_.notify_all();
469
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 99 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
 99 if (reactor_running_ && !reactor_interrupted_)
470 {
471 ✗ reactor_interrupted_ = true;
472 ✗ lock.unlock();
473 ✗ interrupt_reactor();
474 }
475 99 }
476 258580 }
477
478 void
479 15066 select_scheduler::
480 interrupt_reactor() const
481 {
482 15066 char byte = 1;
483
1/1
✓ Branch 1 taken 15066 times.
 15066 [[maybe_unused]] auto r = ::write(pipe_fds_[1], &byte, 1);
484 15066 }
485
486 void
487 251180 select_scheduler::
488 wake_one_thread_and_unlock(std::unique_lock<std::mutex>& lock) const
489 {
490
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 251180 times.
 251180 if (idle_thread_count_ > 0)
491 {
492 // Idle worker exists - wake it via condvar
493 ✗ wakeup_event_.notify_one();
494 ✗ lock.unlock();
495 }
496
4/4
✓ Branch 0 taken 3769 times.
✓ Branch 1 taken 247411 times.
✓ Branch 2 taken 3763 times.
✓ Branch 3 taken 6 times.
 251180 else if (reactor_running_ && !reactor_interrupted_)
497 {
498 // No idle workers but reactor is running - interrupt it
499 3763 reactor_interrupted_ = true;
500 3763 lock.unlock();
501 3763 interrupt_reactor();
502 }
503 else
504 {
505 // No one to wake
506 247417 lock.unlock();
507 }
508 251180 }
509
510 struct work_guard
511 {
512 select_scheduler const* self;
513 258410 ~work_guard() { self->work_finished(); }
514 };
515
516 long
517 10946 select_scheduler::
518 calculate_timeout(long requested_timeout_us) const
519 {
520
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 10946 times.
 10946 if (requested_timeout_us == 0)
521 ✗ return 0;
522
523 10946 auto nearest = timer_svc_->nearest_expiry();
524
3/3
✓ Branch 2 taken 10946 times.
✓ Branch 4 taken 35 times.
✓ Branch 5 taken 10911 times.
 10946 if (nearest == timer_service::time_point::max())
525 35 return requested_timeout_us;
526
527 10911 auto now = std::chrono::steady_clock::now();
528
3/3
✓ Branch 1 taken 10911 times.
✓ Branch 4 taken 54 times.
✓ Branch 5 taken 10857 times.
 10911 if (nearest <= now)
529 54 return 0;
530
531
1/1
✓ Branch 1 taken 10857 times.
 10857 auto timer_timeout_us = std::chrono::duration_cast<std::chrono::microseconds>(
532
1/1
✓ Branch 1 taken 10857 times.
 21714 nearest - now).count();
533
534
1/2
✓ Branch 0 taken 10857 times.
✗ Branch 1 not taken.
 10857 if (requested_timeout_us < 0)
535 10857 return static_cast<long>(timer_timeout_us);
536
537 ✗ return static_cast<long>((std::min)(
538 ✗ static_cast<long long>(requested_timeout_us),
539 ✗ static_cast<long long>(timer_timeout_us)));
540 }
541
542 void
543 137712 select_scheduler::
544 run_reactor(std::unique_lock<std::mutex>& lock)
545 {
546 // Calculate timeout considering timers, use 0 if interrupted
547
3/3
✓ Branch 0 taken 126766 times.
✓ Branch 1 taken 10946 times.
✓ Branch 3 taken 10946 times.
 137712 long effective_timeout_us = reactor_interrupted_ ? 0 : calculate_timeout(-1);
548
549 // Build fd_sets from registered_fds_
550 fd_set read_fds, write_fds, except_fds;
551
2/2
✓ Branch 0 taken 2203392 times.
✓ Branch 1 taken 137712 times.
 2341104 FD_ZERO(&read_fds);
552
2/2
✓ Branch 0 taken 2203392 times.
✓ Branch 1 taken 137712 times.
 2341104 FD_ZERO(&write_fds);
553
2/2
✓ Branch 0 taken 2203392 times.
✓ Branch 1 taken 137712 times.
 2341104 FD_ZERO(&except_fds);
554
555 // Always include the interrupt pipe
556 137712 FD_SET(pipe_fds_[0], &read_fds);
557 137712 int nfds = pipe_fds_[0];
558
559 // Add registered fds
560
2/2
✓ Branch 7 taken 17785 times.
✓ Branch 8 taken 137712 times.
 155497 for (auto& [fd, state] : registered_fds_)
561 {
562
2/2
✓ Branch 0 taken 14230 times.
✓ Branch 1 taken 3555 times.
 17785 if (state.read_op)
563 14230 FD_SET(fd, &read_fds);
564
2/2
✓ Branch 0 taken 3555 times.
✓ Branch 1 taken 14230 times.
 17785 if (state.write_op)
565 {
566 3555 FD_SET(fd, &write_fds);
567 // Also monitor for errors on connect operations
568 3555 FD_SET(fd, &except_fds);
569 }
570
2/2
✓ Branch 0 taken 14234 times.
✓ Branch 1 taken 3551 times.
 17785 if (fd > nfds)
571 14234 nfds = fd;
572 }
573
574 // Convert timeout to timeval
575 struct timeval tv;
576 137712 struct timeval* tv_ptr = nullptr;
577
2/2
✓ Branch 0 taken 137677 times.
✓ Branch 1 taken 35 times.
 137712 if (effective_timeout_us >= 0)
578 {
579 137677 tv.tv_sec = effective_timeout_us / 1000000;
580 137677 tv.tv_usec = effective_timeout_us % 1000000;
581 137677 tv_ptr = &tv;
582 }
583
584
1/1
✓ Branch 1 taken 137712 times.
 137712 lock.unlock();
585
586
1/1
✓ Branch 1 taken 137712 times.
 137712 int ready = ::select(nfds + 1, &read_fds, &write_fds, &except_fds, tv_ptr);
587 137712 int saved_errno = errno;
588
589 // Process timers outside the lock
590
1/1
✓ Branch 1 taken 137712 times.
 137712 timer_svc_->process_expired();
591
592
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 137712 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
 137712 if (ready < 0 && saved_errno != EINTR)
593 ✗ detail::throw_system_error(make_err(saved_errno), "select");
594
595 // Re-acquire lock before modifying completed_ops_
596
1/1
✓ Branch 1 taken 137712 times.
 137712 lock.lock();
597
598 // Drain the interrupt pipe if readable
599
3/4
✓ Branch 0 taken 11280 times.
✓ Branch 1 taken 126432 times.
✓ Branch 2 taken 11280 times.
✗ Branch 3 not taken.
 137712 if (ready > 0 && FD_ISSET(pipe_fds_[0], &read_fds))
600 {
601 char buf[256];
602
3/3
✓ Branch 1 taken 22560 times.
✓ Branch 3 taken 11280 times.
✓ Branch 4 taken 11280 times.
 22560 while (::read(pipe_fds_[0], buf, sizeof(buf)) > 0) {}
603 }
604
605 // Process I/O completions
606 137712 int completions_queued = 0;
607
2/2
✓ Branch 0 taken 11280 times.
✓ Branch 1 taken 126432 times.
 137712 if (ready > 0)
608 {
609 // Iterate over registered fds (copy keys to avoid iterator invalidation)
610 11280 std::vector<int> fds_to_check;
611
1/1
✓ Branch 2 taken 11280 times.
 11280 fds_to_check.reserve(registered_fds_.size());
612
2/2
✓ Branch 7 taken 14264 times.
✓ Branch 8 taken 11280 times.
 25544 for (auto& [fd, state] : registered_fds_)
613
1/1
✓ Branch 1 taken 14264 times.
 14264 fds_to_check.push_back(fd);
614
615
2/2
✓ Branch 5 taken 14264 times.
✓ Branch 6 taken 11280 times.
 25544 for (int fd : fds_to_check)
616 {
617
1/1
✓ Branch 1 taken 14264 times.
 14264 auto it = registered_fds_.find(fd);
618
1/2
✗ Branch 2 not taken.
✓ Branch 3 taken 14264 times.
 14264 if (it == registered_fds_.end())
619 ✗ continue;
620
621 14264 auto& state = it->second;
622
623 // Check for errors (especially for connect operations)
624 14264 bool has_error = FD_ISSET(fd, &except_fds);
625
626 // Process read readiness
627
5/6
✓ Branch 0 taken 10709 times.
✓ Branch 1 taken 3555 times.
✓ Branch 2 taken 7034 times.
✓ Branch 3 taken 3675 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 7034 times.
 14264 if (state.read_op && (FD_ISSET(fd, &read_fds) || has_error))
628 {
629 3675 auto* op = state.read_op;
630 // Claim the op by exchanging to unregistered. Both registering and
631 // registered states mean the op is ours to complete.
632 3675 auto prev = op->registered.exchange(
633 select_registration_state::unregistered, std::memory_order_acq_rel);
634
1/2
✓ Branch 0 taken 3675 times.
✗ Branch 1 not taken.
 3675 if (prev != select_registration_state::unregistered)
635 {
636 3675 state.read_op = nullptr;
637
638
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 3675 times.
 3675 if (has_error)
639 {
640 ✗ int errn = 0;
641 ✗ socklen_t len = sizeof(errn);
642 ✗ if (::getsockopt(fd, SOL_SOCKET, SO_ERROR, &errn, &len) < 0)
643 ✗ errn = errno;
644 ✗ if (errn == 0)
645 ✗ errn = EIO;
646 ✗ op->complete(errn, 0);
647 }
648 else
649 {
650 3675 op->perform_io();
651 }
652
653 3675 completed_ops_.push(op);
654 3675 ++completions_queued;
655 }
656 }
657
658 // Process write readiness
659
3/6
✓ Branch 0 taken 3555 times.
✓ Branch 1 taken 10709 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 3555 times.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
 14264 if (state.write_op && (FD_ISSET(fd, &write_fds) || has_error))
660 {
661 3555 auto* op = state.write_op;
662 // Claim the op by exchanging to unregistered. Both registering and
663 // registered states mean the op is ours to complete.
664 3555 auto prev = op->registered.exchange(
665 select_registration_state::unregistered, std::memory_order_acq_rel);
666
1/2
✓ Branch 0 taken 3555 times.
✗ Branch 1 not taken.
 3555 if (prev != select_registration_state::unregistered)
667 {
668 3555 state.write_op = nullptr;
669
670
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 3555 times.
 3555 if (has_error)
671 {
672 ✗ int errn = 0;
673 ✗ socklen_t len = sizeof(errn);
674 ✗ if (::getsockopt(fd, SOL_SOCKET, SO_ERROR, &errn, &len) < 0)
675 ✗ errn = errno;
676 ✗ if (errn == 0)
677 ✗ errn = EIO;
678 ✗ op->complete(errn, 0);
679 }
680 else
681 {
682 3555 op->perform_io();
683 }
684
685 3555 completed_ops_.push(op);
686 3555 ++completions_queued;
687 }
688 }
689
690 // Clean up empty entries
691
3/4
✓ Branch 0 taken 7230 times.
✓ Branch 1 taken 7034 times.
✓ Branch 2 taken 7230 times.
✗ Branch 3 not taken.
 14264 if (!state.read_op && !state.write_op)
692
1/1
✓ Branch 1 taken 7230 times.
 7230 registered_fds_.erase(it);
693 }
694 11280 }
695
696
2/2
✓ Branch 0 taken 3679 times.
✓ Branch 1 taken 134033 times.
 137712 if (completions_queued > 0)
697 {
698
2/2
✓ Branch 0 taken 128 times.
✓ Branch 1 taken 3551 times.
 3679 if (completions_queued == 1)
699 128 wakeup_event_.notify_one();
700 else
701 3551 wakeup_event_.notify_all();
702 }
703 137712 }
704
705 std::size_t
706 258506 select_scheduler::
707 do_one(long timeout_us)
708 {
709
1/1
✓ Branch 1 taken 258506 times.
 258506 std::unique_lock lock(mutex_);
710
711 for (;;)
712 {
713
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 396218 times.
 396218 if (stopped_.load(std::memory_order_acquire))
714 ✗ return 0;
715
716 396218 scheduler_op* op = completed_ops_.pop();
717
718
2/2
✓ Branch 0 taken 137808 times.
✓ Branch 1 taken 258410 times.
 396218 if (op == &task_op_)
719 {
720 137808 bool more_handlers = !completed_ops_.empty();
721
722
2/2
✓ Branch 0 taken 11042 times.
✓ Branch 1 taken 126766 times.
 137808 if (!more_handlers)
723 {
724
2/2
✓ Branch 1 taken 96 times.
✓ Branch 2 taken 10946 times.
 22084 if (outstanding_work_.load(std::memory_order_acquire) == 0)
725 {
726 96 completed_ops_.push(&task_op_);
727 96 return 0;
728 }
729
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 10946 times.
 10946 if (timeout_us == 0)
730 {
731 ✗ completed_ops_.push(&task_op_);
732 ✗ return 0;
733 }
734 }
735
736
3/4
✓ Branch 0 taken 10946 times.
✓ Branch 1 taken 126766 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 10946 times.
 137712 reactor_interrupted_ = more_handlers || timeout_us == 0;
737 137712 reactor_running_ = true;
738
739
3/4
✓ Branch 0 taken 126766 times.
✓ Branch 1 taken 10946 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 126766 times.
 137712 if (more_handlers && idle_thread_count_ > 0)
740 ✗ wakeup_event_.notify_one();
741
742
1/1
✓ Branch 1 taken 137712 times.
 137712 run_reactor(lock);
743
744 137712 reactor_running_ = false;
745 137712 completed_ops_.push(&task_op_);
746 137712 continue;
747 137712 }
748
749
1/2
✓ Branch 0 taken 258410 times.
✗ Branch 1 not taken.
 258410 if (op != nullptr)
750 {
751
1/1
✓ Branch 1 taken 258410 times.
 258410 lock.unlock();
752 258410 work_guard g{this};
753
1/1
✓ Branch 1 taken 258410 times.
 258410 (*op)();
754 258410 return 1;
755 258410 }
756
757 ✗ if (outstanding_work_.load(std::memory_order_acquire) == 0)
758 ✗ return 0;
759
760 ✗ if (timeout_us == 0)
761 ✗ return 0;
762
763 ✗ ++idle_thread_count_;
764 ✗ if (timeout_us < 0)
765 ✗ wakeup_event_.wait(lock);
766 else
767 ✗ wakeup_event_.wait_for(lock, std::chrono::microseconds(timeout_us));
768 ✗ --idle_thread_count_;
769 137712 }
770 258506 }
771
772 } // namespace boost::corosio::detail
773
774 #endif // BOOST_COROSIO_HAS_SELECT
775