1 /*
2     Copyright (c) 2005-2021 Intel Corporation
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 #include "oneapi/tbb/cache_aligned_allocator.h"
18 #include "oneapi/tbb/mutex.h"
19 
20 #include "rml_tbb.h"
21 #include "rml_thread_monitor.h"
22 
23 #include "scheduler_common.h"
24 #include "governor.h"
25 #include "misc.h"
26 
27 #include <atomic>
28 
29 
30 namespace tbb {
31 namespace detail {
32 namespace r1 {
33 namespace rml {
34 
35 using rml::internal::thread_monitor;
36 typedef thread_monitor::handle_type thread_handle;
37 
38 class private_server;
39 
40 class private_worker: no_copy {
41 private:
42     //! State in finite-state machine that controls the worker.
43     /** State diagram:
44         init --> starting --> normal
45           |         |           |
46           |         V           |
47           \------> quit <------/
48       */
49     enum state_t {
50         //! *this is initialized
51         st_init,
52         //! *this has associated thread that is starting up.
53         st_starting,
54         //! Associated thread is doing normal life sequence.
55         st_normal,
56         //! Associated thread has ended normal life sequence and promises to never touch *this again.
57         st_quit
58     };
59     std::atomic<state_t> my_state;
60 
61     //! Associated server
62     private_server& my_server;
63 
64     //! Associated client
65     tbb_client& my_client;
66 
67     //! index used for avoiding the 64K aliasing problem
68     const std::size_t my_index;
69 
70     //! Monitor for sleeping when there is no work to do.
71     /** The invariant that holds for sleeping workers is:
72         "my_slack<=0 && my_state==st_normal && I am on server's list of asleep threads" */
73     thread_monitor my_thread_monitor;
74 
75     //! Handle of the OS thread associated with this worker
76     thread_handle my_handle;
77 
78     //! Link for list of workers that are sleeping or have no associated thread.
79     private_worker* my_next;
80 
81     friend class private_server;
82 
83     //! Actions executed by the associated thread
84     void run() noexcept;
85 
86     //! Wake up associated thread (or launch a thread if there is none)
87     void wake_or_launch();
88 
89     //! Called by a thread (usually not the associated thread) to commence termination.
90     void start_shutdown();
91 
92     static __RML_DECL_THREAD_ROUTINE thread_routine( void* arg );
93 
94     static void release_handle(thread_handle my_handle, bool join);
95 
96 protected:
private_worker(private_server & server,tbb_client & client,const std::size_t i)97     private_worker( private_server& server, tbb_client& client, const std::size_t i ) :
98         my_state(st_init), my_server(server), my_client(client), my_index(i),
99         my_thread_monitor(), my_handle(), my_next()
100     {}
101 };
102 
103 static const std::size_t cache_line_size = tbb::detail::max_nfs_size;
104 
105 #if _MSC_VER && !defined(__INTEL_COMPILER)
106     // Suppress overzealous compiler warnings about uninstantiable class
107     #pragma warning(push)
108     #pragma warning(disable:4510 4610)
109 #endif
110 class padded_private_worker: public private_worker {
111     char pad[cache_line_size - sizeof(private_worker)%cache_line_size];
112 public:
padded_private_worker(private_server & server,tbb_client & client,const std::size_t i)113     padded_private_worker( private_server& server, tbb_client& client, const std::size_t i )
114     : private_worker(server,client,i) { suppress_unused_warning(pad); }
115 };
116 #if _MSC_VER && !defined(__INTEL_COMPILER)
117     #pragma warning(pop)
118 #endif
119 
120 class private_server: public tbb_server, no_copy {
121 private:
122     tbb_client& my_client;
123     //! Maximum number of threads to be created.
124     /** Threads are created lazily, so maximum might not actually be reached. */
125     const tbb_client::size_type my_n_thread;
126 
127     //! Stack size for each thread. */
128     const std::size_t my_stack_size;
129 
130     //! Number of jobs that could use their associated thread minus number of active threads.
131     /** If negative, indicates oversubscription.
132         If positive, indicates that more threads should run.
133         Can be lowered asynchronously, but must be raised only while holding my_asleep_list_mutex,
134         because raising it impacts the invariant for sleeping threads. */
135     std::atomic<int> my_slack;
136 
137     //! Counter used to determine when to delete this.
138     std::atomic<int> my_ref_count;
139 
140     padded_private_worker* my_thread_array;
141 
142     //! List of workers that are asleep or committed to sleeping until notified by another thread.
143     std::atomic<private_worker*> my_asleep_list_root;
144 
145     //! Protects my_asleep_list_root
146     typedef mutex asleep_list_mutex_type;
147     asleep_list_mutex_type my_asleep_list_mutex;
148 
149 #if TBB_USE_ASSERT
150     std::atomic<int> my_net_slack_requests;
151 #endif /* TBB_USE_ASSERT */
152 
153     //! Wake up to two sleeping workers, if there are any sleeping.
154     /** The call is used to propagate a chain reaction where each thread wakes up two threads,
155         which in turn each wake up two threads, etc. */
propagate_chain_reaction()156     void propagate_chain_reaction() {
157         // First test of a double-check idiom.  Second test is inside wake_some(0).
158         if( my_asleep_list_root.load(std::memory_order_relaxed) )
159             wake_some(0);
160     }
161 
162     //! Try to add t to list of sleeping workers
163     bool try_insert_in_asleep_list( private_worker& t );
164 
165     //! Equivalent of adding additional_slack to my_slack and waking up to 2 threads if my_slack permits.
166     void wake_some( int additional_slack );
167 
168     virtual ~private_server();
169 
remove_server_ref()170     void remove_server_ref() {
171         if( --my_ref_count==0 ) {
172             my_client.acknowledge_close_connection();
173             this->~private_server();
174             tbb::cache_aligned_allocator<private_server>().deallocate( this, 1 );
175         }
176     }
177 
178     friend class private_worker;
179 public:
180     private_server( tbb_client& client );
181 
version() const182     version_type version() const override {
183         return 0;
184     }
185 
request_close_connection(bool)186     void request_close_connection( bool /*exiting*/ ) override {
187         for( std::size_t i=0; i<my_n_thread; ++i )
188             my_thread_array[i].start_shutdown();
189         remove_server_ref();
190     }
191 
yield()192     void yield() override { d0::yield(); }
193 
independent_thread_number_changed(int)194     void independent_thread_number_changed( int ) override {__TBB_ASSERT(false,NULL);}
195 
default_concurrency() const196     unsigned default_concurrency() const override { return governor::default_num_threads() - 1; }
197 
198     void adjust_job_count_estimate( int delta ) override;
199 
200 #if _WIN32 || _WIN64
register_external_thread(::rml::server::execution_resource_t &)201     void register_external_thread ( ::rml::server::execution_resource_t& ) override {}
unregister_external_thread(::rml::server::execution_resource_t)202     void unregister_external_thread ( ::rml::server::execution_resource_t ) override {}
203 #endif /* _WIN32||_WIN64 */
204 };
205 
206 //------------------------------------------------------------------------
207 // Methods of private_worker
208 //------------------------------------------------------------------------
209 #if _MSC_VER && !defined(__INTEL_COMPILER)
210     // Suppress overzealous compiler warnings about an initialized variable 'sink_for_alloca' not referenced
211     #pragma warning(push)
212     #pragma warning(disable:4189)
213 #endif
214 #if __MINGW32__ && __GNUC__==4 &&__GNUC_MINOR__>=2 && !__MINGW64__
215 // ensure that stack is properly aligned for TBB threads
216 __attribute__((force_align_arg_pointer))
217 #endif
thread_routine(void * arg)218 __RML_DECL_THREAD_ROUTINE private_worker::thread_routine( void* arg ) {
219     private_worker* self = static_cast<private_worker*>(arg);
220     AVOID_64K_ALIASING( self->my_index );
221     self->run();
222     return 0;
223 }
224 #if _MSC_VER && !defined(__INTEL_COMPILER)
225     #pragma warning(pop)
226 #endif
227 
release_handle(thread_handle handle,bool join)228 void private_worker::release_handle(thread_handle handle, bool join) {
229     if (join)
230         thread_monitor::join(handle);
231     else
232         thread_monitor::detach_thread(handle);
233 }
234 
start_shutdown()235 void private_worker::start_shutdown() {
236     state_t expected_state = my_state.load(std::memory_order_acquire);
237     __TBB_ASSERT( expected_state!=st_quit, NULL );
238 
239     while( !my_state.compare_exchange_strong( expected_state, st_quit ) );
240 
241     if( expected_state==st_normal || expected_state==st_starting ) {
242         // May have invalidated invariant for sleeping, so wake up the thread.
243         // Note that the notify() here occurs without maintaining invariants for my_slack.
244         // It does not matter, because my_state==st_quit overrides checking of my_slack.
245         my_thread_monitor.notify();
246         // Do not need release handle in st_init state,
247         // because in this case the thread wasn't started yet.
248         // For st_starting release is done at launch site.
249         if (expected_state==st_normal)
250             release_handle(my_handle, governor::does_client_join_workers(my_client));
251     } else if( expected_state==st_init ) {
252         // Perform action that otherwise would be performed by associated thread when it quits.
253         my_server.remove_server_ref();
254     }
255 }
256 
run()257 void private_worker::run() noexcept {
258     my_server.propagate_chain_reaction();
259 
260     // Transiting to st_normal here would require setting my_handle,
261     // which would create race with the launching thread and
262     // complications in handle management on Windows.
263 
264     ::rml::job& j = *my_client.create_one_job();
265     while( my_state.load(std::memory_order_acquire)!=st_quit ) {
266         if( my_server.my_slack.load(std::memory_order_acquire)>=0 ) {
267             my_client.process(j);
268         } else {
269             thread_monitor::cookie c;
270             // Prepare to wait
271             my_thread_monitor.prepare_wait(c);
272             // Check/set the invariant for sleeping
273             if( my_state.load(std::memory_order_acquire)!=st_quit && my_server.try_insert_in_asleep_list(*this) ) {
274                 my_thread_monitor.commit_wait(c);
275                 __TBB_ASSERT( my_state==st_quit || !my_next, "Thread monitor missed a spurious wakeup?" );
276                 my_server.propagate_chain_reaction();
277             } else {
278                 // Invariant broken
279                 my_thread_monitor.cancel_wait();
280             }
281         }
282     }
283     my_client.cleanup(j);
284 
285     ++my_server.my_slack;
286     my_server.remove_server_ref();
287 }
288 
wake_or_launch()289 inline void private_worker::wake_or_launch() {
290     state_t expected_state = st_init;
291     if( my_state.compare_exchange_strong( expected_state, st_starting ) ) {
292         // after this point, remove_server_ref() must be done by created thread
293 #if __TBB_USE_WINAPI
294         my_handle = thread_monitor::launch( thread_routine, this, my_server.my_stack_size, &this->my_index );
295 #elif __TBB_USE_POSIX
296         {
297         affinity_helper fpa;
298         fpa.protect_affinity_mask( /*restore_process_mask=*/true );
299         my_handle = thread_monitor::launch( thread_routine, this, my_server.my_stack_size );
300         // Implicit destruction of fpa resets original affinity mask.
301         }
302 #endif /* __TBB_USE_POSIX */
303         expected_state = st_starting;
304         if ( !my_state.compare_exchange_strong( expected_state, st_normal ) ) {
305             // Do shutdown during startup. my_handle can't be released
306             // by start_shutdown, because my_handle value might be not set yet
307             // at time of transition from st_starting to st_quit.
308             __TBB_ASSERT( expected_state==st_quit, NULL );
309             release_handle(my_handle, governor::does_client_join_workers(my_client));
310         }
311     }
312     else {
313         __TBB_ASSERT( !my_next, "Should not wake a thread while it's still in asleep list" );
314         my_thread_monitor.notify();
315     }
316 }
317 
318 //------------------------------------------------------------------------
319 // Methods of private_server
320 //------------------------------------------------------------------------
private_server(tbb_client & client)321 private_server::private_server( tbb_client& client ) :
322     my_client(client),
323     my_n_thread(client.max_job_count()),
324     my_stack_size(client.min_stack_size()),
325     my_slack(0),
326     my_ref_count(my_n_thread+1),
327     my_thread_array(NULL),
328     my_asleep_list_root(NULL)
329 #if TBB_USE_ASSERT
330     , my_net_slack_requests(0)
331 #endif /* TBB_USE_ASSERT */
332 {
333     my_thread_array = tbb::cache_aligned_allocator<padded_private_worker>().allocate( my_n_thread );
334     for( std::size_t i=0; i<my_n_thread; ++i ) {
335         private_worker* t = new( &my_thread_array[i] ) padded_private_worker( *this, client, i );
336         t->my_next = my_asleep_list_root.load(std::memory_order_relaxed);
337         my_asleep_list_root.store(t, std::memory_order_relaxed);
338     }
339 }
340 
~private_server()341 private_server::~private_server() {
342     __TBB_ASSERT( my_net_slack_requests==0, NULL );
343     for( std::size_t i=my_n_thread; i--; )
344         my_thread_array[i].~padded_private_worker();
345     tbb::cache_aligned_allocator<padded_private_worker>().deallocate( my_thread_array, my_n_thread );
346     tbb::detail::poison_pointer( my_thread_array );
347 }
348 
try_insert_in_asleep_list(private_worker & t)349 inline bool private_server::try_insert_in_asleep_list( private_worker& t ) {
350     asleep_list_mutex_type::scoped_lock lock;
351     if( !lock.try_acquire(my_asleep_list_mutex) )
352         return false;
353     // Contribute to slack under lock so that if another takes that unit of slack,
354     // it sees us sleeping on the list and wakes us up.
355     auto expected = my_slack.load(std::memory_order_relaxed);
356     while (expected < 0) {
357         if (my_slack.compare_exchange_strong(expected, expected + 1)) {
358             t.my_next = my_asleep_list_root.load(std::memory_order_relaxed);
359             my_asleep_list_root.store(&t, std::memory_order_relaxed);
360             return true;
361         }
362     }
363 
364     return false;
365 }
366 
wake_some(int additional_slack)367 void private_server::wake_some( int additional_slack ) {
368     __TBB_ASSERT( additional_slack>=0, nullptr );
369     private_worker* wakee[2];
370     private_worker**w = wakee;
371 
372     if (additional_slack) {
373         // Contribute our unused slack to my_slack.
374         my_slack += additional_slack;
375     }
376 
377     int allotted_slack = 0;
378     while (allotted_slack < 2) {
379         // Chain reaction; Try to claim unit of slack
380         int old = my_slack.load(std::memory_order_relaxed);
381         do {
382             if (old <= 0) goto done;
383         } while (!my_slack.compare_exchange_strong(old, old - 1));
384         ++allotted_slack;
385     }
386 done:
387 
388     if (allotted_slack)
389     {
390         asleep_list_mutex_type::scoped_lock lock(my_asleep_list_mutex);
391 
392         while( my_asleep_list_root.load(std::memory_order_relaxed) && w<wakee+2 && allotted_slack) {
393             --allotted_slack;
394             // Pop sleeping worker to combine with claimed unit of slack
395             auto old = my_asleep_list_root.load(std::memory_order_relaxed);
396             my_asleep_list_root.store(old->my_next, std::memory_order_relaxed);
397             *w++ = old;
398         }
399         if(allotted_slack) {
400             // Contribute our unused slack to my_slack.
401             my_slack += allotted_slack;
402         }
403     }
404     while( w>wakee ) {
405         private_worker* ww = *--w;
406         ww->my_next = nullptr;
407         ww->wake_or_launch();
408     }
409 }
410 
adjust_job_count_estimate(int delta)411 void private_server::adjust_job_count_estimate( int delta ) {
412 #if TBB_USE_ASSERT
413     my_net_slack_requests+=delta;
414 #endif /* TBB_USE_ASSERT */
415     if( delta<0 ) {
416         my_slack+=delta;
417     } else if( delta>0 ) {
418         wake_some( delta );
419     }
420 }
421 
422 //! Factory method called from task.cpp to create a private_server.
make_private_server(tbb_client & client)423 tbb_server* make_private_server( tbb_client& client ) {
424     return new( tbb::cache_aligned_allocator<private_server>().allocate(1) ) private_server(client);
425 }
426 
427 } // namespace rml
428 } // namespace r1
429 } // namespace detail
430 } // namespace tbb
431 
432