summaryrefslogtreecommitdiff
path: root/rts/win32/MIOManager.c
blob: 06b28bca2ba4e6cdb6f1ba101bbfe5b9a7927255 (plain)
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
616
/* MIOManager.c
 *
 * Non-blocking / asynchronous I/O for Win32.
 *
 * (c) sof, 2002-2003.
 *
 * NOTE: This is the MIO manager, only used for --io-manager=posix.
 *       For the WINIO manager see base in the GHC.Event modules.
 */

#if !defined(THREADED_RTS)

#include "Rts.h"
#include "RtsUtils.h"
#include "MIOManager.h"
#include "WorkQueue.h"
#include "ConsoleHandler.h"
#include <stdio.h>
#include <stdlib.h>
#include <io.h>
#include <winsock.h>
#include <process.h>
#include <errno.h>

/*
 * Internal state maintained by the IO manager.
 */
typedef struct IOManagerState {
    Mutex            manLock;
    WorkQueue*       workQueue;
    int              queueSize;
    int              numWorkers;
    int              workersIdle;
    HANDLE           hExitEvent;
    unsigned int     requestID;
    /* fields for keeping track of active WorkItems */
    Mutex            active_work_lock;
    WorkItem*        active_work_items;
    UINT             sleepResolution;
} IOManagerState;

/* ToDo: wrap up this state via a IOManager handle instead? */
static IOManagerState* ioMan;

static void RegisterWorkItem  ( IOManagerState* iom, WorkItem* wi);
static void DeregisterWorkItem( IOManagerState* iom, WorkItem* wi);

/*
 * The routine executed by each worker thread.
 */
static
unsigned
WINAPI
IOWorkerProc(PVOID param)
{
    HANDLE  hWaits[2];
    DWORD   rc;
    IOManagerState* iom = (IOManagerState*)param;
    WorkQueue* pq = iom->workQueue;
    WorkItem*  work;
    int        len = 0, fd = 0;
    DWORD      errCode = 0;
    void*      complData;

    hWaits[0] = (HANDLE)iom->hExitEvent;
    hWaits[1] = GetWorkQueueHandle(pq);

    while (1) {
        // The error code is communicated back on completion of request; reset.
        errCode = 0;

        OS_ACQUIRE_LOCK(&iom->manLock);
        /* Signal that the worker is idle.
         *
         * 'workersIdle' is used when determining whether or not to
         * increase the worker thread pool when adding a new request.
         * (see addIORequest().)
         */
        iom->workersIdle++;
        OS_RELEASE_LOCK(&iom->manLock);

        /*
         * A possible future refinement is to make long-term idle threads
         * wake up and decide to shut down should the number of idle threads
         * be above some threshold.
         *
         */
        rc = WaitForMultipleObjects( 2, hWaits, false, INFINITE );

        if (rc == WAIT_OBJECT_0) {
            // we received the exit event
            OS_ACQUIRE_LOCK(&iom->manLock);
            ioMan->numWorkers--;
            OS_RELEASE_LOCK(&iom->manLock);
            return 0;
        }

        OS_ACQUIRE_LOCK(&iom->manLock);
        /* Signal that the thread is 'non-idle' and about to consume
         * a work item.
         */
        iom->workersIdle--;
        iom->queueSize--;
        OS_RELEASE_LOCK(&iom->manLock);

        if ( rc == (WAIT_OBJECT_0 + 1) ) {
            /* work item available, fetch it. */
            if (FetchWork(pq,(void**)&work)) {
                work->abandonOp = 0;
                RegisterWorkItem(iom,work);
                if ( work->workKind & WORKER_READ ) {
                    if ( work->workKind & WORKER_FOR_SOCKET ) {
                        len = recv(work->workData.ioData.fd,
                                   work->workData.ioData.buf,
                                   work->workData.ioData.len,
                                   0);
                        if (len == SOCKET_ERROR) {
                            errCode = WSAGetLastError();
                        }
                    } else {
                        while (1) {
                        // Do the read(), with extra-special handling for Ctrl+C
                        len = read(work->workData.ioData.fd,
                                   work->workData.ioData.buf,
                                   work->workData.ioData.len);
                        if ( len == 0 && work->workData.ioData.len != 0 ) {
                            /* Given the following scenario:
                             * - a console handler has been registered
                             *   that handles Ctrl+C events.
                             * - we've not tweaked the 'console mode'
                             *   settings to turn on
                             *   ENABLE_PROCESSED_INPUT.
                             * - we're blocked waiting on input from
                                 standard input.
                             * - the user hits Ctrl+C.
                             *
                             * The OS will invoke the console handler
                             * (in a separate OS thread), and the
                             * above read() (i.e., under the hood, a
                             * ReadFile() op) returns 0, with the
                             * error set to
                             * ERROR_OPERATION_ABORTED. We don't want
                             * to percolate this error condition back
                             * to the Haskell user.  Do this by
                             * waiting for the completion of the
                             * Haskell console handler. If upon
                             * completion of the console handler
                             * routine, the Haskell thread that issued
                             * the request is found to have been
                             * thrown an exception, the worker
                             * abandons the request (since that's what
                             * the Haskell thread has done.) If the
                             * Haskell thread hasn't been interrupted,
                             * the worker retries the read request as
                             * if nothing happened.
                             */
                            if ( (GetLastError()) == ERROR_OPERATION_ABORTED ) {
                                /* For now, only abort when dealing
                                 * with the standard input handle.
                                 * i.e., for all others, an error is
                                 * raised.
                                 */
                                HANDLE h =
                                    (HANDLE)GetStdHandle(STD_INPUT_HANDLE);
                                int iofd = work->workData.ioData.fd;
                                if ( _get_osfhandle(iofd) == (intptr_t)h ) {
                                    if (rts_waitConsoleHandlerCompletion()) {
                                        /* If the Scheduler has set
                                         * work->abandonOp, the
                                         * Haskell thread has been
                                         * thrown an exception (=> the
                                         * worker must abandon this
                                         * request.)  We test for this
                                         * below before invoking the
                                         * on-completion routine.
                                         */
                                        if (work->abandonOp) {
                                            break;
                                        } else {
                                            continue;
                                        }
                                    }
                                } else {
                                    break; /* Treat it like an error */
                                }
                            } else {
                                break;
                            }
                        } else {
                            break;
                        }
                        }
                        if (len == -1) { errCode = errno; }
                    }
                    complData = work->workData.ioData.buf;
                    fd = work->workData.ioData.fd;
                } else if ( work->workKind & WORKER_WRITE ) {
                    if ( work->workKind & WORKER_FOR_SOCKET ) {
                        len = send(work->workData.ioData.fd,
                                   work->workData.ioData.buf,
                                   work->workData.ioData.len,
                                   0);
                        if (len == SOCKET_ERROR) {
                            errCode = WSAGetLastError();
                        }
                    } else {
                        len = write(work->workData.ioData.fd,
                                    work->workData.ioData.buf,
                                    work->workData.ioData.len);
                        if (len == -1) {
                            errCode = errno;
                            // write() gets errno wrong for
                            // ERROR_NO_DATA, we have to fix it here:
                            if (errCode == EINVAL &&
                                GetLastError() == ERROR_NO_DATA) {
                                errCode = EPIPE;
                            }
                        }
                    }
                    complData = work->workData.ioData.buf;
                    fd = work->workData.ioData.fd;
                } else if ( work->workKind & WORKER_DELAY ) {
                    /* Approximate implementation of threadDelay;
                     *
                     * Note: Sleep() is in milliseconds, not micros.
                     *
                     * MSDN says of Sleep:
                     *   If dwMilliseconds is greater than one tick
                     *   but less than two, the wait can be anywhere
                     *   between one and two ticks, and so on.
                     *
                     * so we need to add (milliseconds-per-tick - 1)
                     * to the amount of time we sleep for.
                     *
                     * test ThreadDelay001 fails if we get this wrong.
                     */
                    Sleep(((work->workData.delayData.usecs + 999) / 1000)
                            + iom->sleepResolution - 1);
                    len = work->workData.delayData.usecs;
                    complData = NULL;
                    fd = 0;
                    errCode = 0;
                } else if ( work->workKind & WORKER_DO_PROC ) {
                    // perform operation/proc on behalf of Haskell thread.
                    if (work->workData.procData.proc) {
                        // The procedure is assumed to encode result +
                        // success/failure via its param.
                        void* param = work->workData.procData.param;
                        errCode=work->workData.procData.proc(param);
                    } else {
                        errCode=1;
                    }
                    complData = work->workData.procData.param;
                } else {
                    fprintf(stderr, "unknown work request type (%d), "
                                    "ignoring.\n", work->workKind);
                    fflush(stderr);
                    continue;
                }
                if (!work->abandonOp) {
                    work->onCompletion(work->requestID,
                                       fd,
                                       len,
                                       complData,
                                       errCode);
                }
                // Free the WorkItem
                DeregisterWorkItem(iom,work);
                stgFree(work);
            } else {
                fprintf(stderr, "unable to fetch work; fatal.\n");
                fflush(stderr);
                OS_ACQUIRE_LOCK(&iom->manLock);
                ioMan->numWorkers--;
                OS_RELEASE_LOCK(&iom->manLock);
                return 1;
            }
        } else {
            fprintf(stderr, "waiting failed (%lu); fatal.\n", rc);
            fflush(stderr);
            OS_ACQUIRE_LOCK(&iom->manLock);
            ioMan->numWorkers--;
            OS_RELEASE_LOCK(&iom->manLock);
            return 1;
        }
    }
    return 0;
}

static
bool
NewIOWorkerThread(IOManagerState* iom)
{
    unsigned threadId;
    return ( 0 != _beginthreadex(NULL,
                                 0,
                                 IOWorkerProc,
                                 (LPVOID)iom,
                                 0,
                                 &threadId) );
}

bool
StartIOManager(void)
{
    HANDLE hExit;
    WorkQueue* wq;
    UINT sleepResolution;
    TIMECAPS timecaps;
    MMRESULT mmresult;

    mmresult = timeGetDevCaps(&timecaps, sizeof(timecaps));
    if (mmresult != MMSYSERR_NOERROR) {
        return false;
    }
    sleepResolution = timecaps.wPeriodMin;
    mmresult = timeBeginPeriod(sleepResolution);
    if (mmresult != MMSYSERR_NOERROR) {
        return false;
    }

    wq = NewWorkQueue();
    if ( !wq ) return false;

    ioMan = (IOManagerState*)stgMallocBytes(sizeof(IOManagerState), "StartIOManager");

    if (!ioMan) {
        FreeWorkQueue(wq);
        return false;
    }

    /* A manual-reset event */
    hExit = CreateEvent ( NULL, true, false, NULL );
    if ( !hExit ) {
        FreeWorkQueue(wq);
        stgFree(ioMan);
        return false;
    }

    ioMan->hExitEvent = hExit;
    OS_INIT_LOCK(&ioMan->manLock);
    ioMan->workQueue   = wq;
    ioMan->numWorkers  = 0;
    ioMan->workersIdle = 0;
    ioMan->queueSize   = 0;
    ioMan->requestID   = 1;
    OS_INIT_LOCK(&ioMan->active_work_lock);
    ioMan->active_work_items = NULL;
    ioMan->sleepResolution = sleepResolution;

    return true;
}

/*
 * Function: depositWorkItem()
 *
 * Local function which deposits a WorkItem onto a work queue,
 * deciding in the process whether or not the thread pool needs
 * to be augmented with another thread to handle the new request.
 *
 */
static
int
depositWorkItem( unsigned int reqID,
                 WorkItem* wItem )
{
    OS_ACQUIRE_LOCK(&ioMan->manLock);

#if 0
    fprintf(stderr, "depositWorkItem: %d/%d\n",
            ioMan->workersIdle, ioMan->numWorkers);
    fflush(stderr);
#endif
    /* A new worker thread is created when there are fewer idle threads
     * than non-consumed queue requests. This ensures that requests will
     * be dealt with in a timely manner.
     *
     * [Long explanation of why the previous thread pool policy lead to
     * trouble]
     *
     * Previously, the thread pool was augmented iff no idle worker threads
     * were available. That strategy runs the risk of repeatedly adding to
     * the request queue without expanding the thread pool to handle this
     * sudden spike in queued requests.
     * [How? Assume workersIdle is 1, and addIORequest() is called. No new
     * thread is created and the request is simply queued. If addIORequest()
     * is called again _before the OS schedules a worker thread to pull the
     * request off the queue_, workersIdle is still 1 and another request is
     * simply added to the queue. Once the worker thread is run, only one
     * request is de-queued, leaving the 2nd request in the queue]
     *
     * Assuming none of the queued requests take an inordinate amount
     * of to complete, the request queue would eventually be
     * drained. But if that's not the case, the later requests will
     * end up languishing in the queue indefinitely. The non-timely
     * handling of requests may cause CH applications to misbehave /
     * hang; bad.
     *
     */
    ioMan->queueSize++;
    if ( (ioMan->workersIdle < ioMan->queueSize) ) {
        /* see if giving up our quantum ferrets out some idle threads.
         */
        OS_RELEASE_LOCK(&ioMan->manLock);
        Sleep(0);
        OS_ACQUIRE_LOCK(&ioMan->manLock);
        if ( (ioMan->workersIdle < ioMan->queueSize) ) {
            /* No, go ahead and create another. */
            ioMan->numWorkers++;
            if (!NewIOWorkerThread(ioMan)) {
                ioMan->numWorkers--;
            }
        }
    }
    OS_RELEASE_LOCK(&ioMan->manLock);

    if (SubmitWork(ioMan->workQueue,wItem)) {
        /* Note: the work item has potentially been consumed by a worker thread
         *       (and freed) at this point, so we cannot use wItem's requestID.
         */
        return reqID;
    } else {
        return 0;
    }
}

/*
 * Function: AddIORequest()
 *
 * Conduit to underlying WorkQueue's SubmitWork(); adds IO
 * request to work queue, deciding whether or not to augment
 * the thread pool in the process.
 */
int
AddIORequest ( int   fd,
               bool  forWriting,
               bool  isSocket,
               HsInt len,
               char* buffer,
               CompletionProc onCompletion)
{
    ASSERT(ioMan);

    WorkItem* wItem    = (WorkItem*)stgMallocBytes(sizeof(WorkItem), "AddIORequest");

    unsigned int reqID = ioMan->requestID++;

    /* Fill in the blanks */
    wItem->workKind     = ( isSocket   ? WORKER_FOR_SOCKET : 0 ) |
                          ( forWriting ? WORKER_WRITE : WORKER_READ );
    wItem->workData.ioData.fd  = fd;
    wItem->workData.ioData.len = len;
    wItem->workData.ioData.buf = buffer;
    wItem->link = NULL;

    wItem->onCompletion        = onCompletion;
    wItem->requestID           = reqID;

    return depositWorkItem(reqID, wItem);
}

/*
 * Function: AddDelayRequest()
 *
 * Like AddIORequest(), but this time adding a delay request to
 * the request queue.
 */
BOOL
AddDelayRequest ( HsInt          usecs,
                  CompletionProc onCompletion)
{
    ASSERT(ioMan);

    WorkItem* wItem = (WorkItem*)stgMallocBytes(sizeof(WorkItem), "AddDelayRequest");

    unsigned int reqID = ioMan->requestID++;

    /* Fill in the blanks */
    wItem->workKind     = WORKER_DELAY;
    wItem->workData.delayData.usecs = usecs;
    wItem->onCompletion = onCompletion;
    wItem->requestID    = reqID;
    wItem->link         = NULL;

    return depositWorkItem(reqID, wItem);
}

/*
 * Function: AddProcRequest()
 *
 * Add an asynchronous procedure request.
 */
BOOL
AddProcRequest ( void* proc,
                 void* param,
                 CompletionProc onCompletion)
{
    ASSERT(ioMan);

    WorkItem* wItem = (WorkItem*)stgMallocBytes(sizeof(WorkItem), "AddProcRequest");
    if (!wItem) return false;

    unsigned int reqID = ioMan->requestID++;

    /* Fill in the blanks */
    wItem->workKind     = WORKER_DO_PROC;
    wItem->workData.procData.proc  = proc;
    wItem->workData.procData.param = param;
    wItem->onCompletion = onCompletion;
    wItem->requestID    = reqID;
    wItem->abandonOp    = 0;
    wItem->link         = NULL;

    return depositWorkItem(reqID, wItem);
}

void ShutdownIOManager ( bool wait_threads )
{
    int num;
    MMRESULT mmresult;

    SetEvent(ioMan->hExitEvent);

    if (wait_threads) {
        /* Wait for all worker threads to die. */
        for (;;) {
            OS_ACQUIRE_LOCK(&ioMan->manLock);
            num = ioMan->numWorkers;
            OS_RELEASE_LOCK(&ioMan->manLock);
            if (num == 0)
                break;
            Sleep(10);
        }
        FreeWorkQueue(ioMan->workQueue);
        CloseHandle(ioMan->hExitEvent);
        OS_CLOSE_LOCK(&ioMan->active_work_lock);
        OS_CLOSE_LOCK(&ioMan->manLock);

        mmresult = timeEndPeriod(ioMan->sleepResolution);
        if (mmresult != MMSYSERR_NOERROR) {
            barf("timeEndPeriod failed");
        }

        stgFree(ioMan);
        ioMan = NULL;
    }
}

/* Keep track of WorkItems currently being serviced. */
static
void
RegisterWorkItem(IOManagerState* ioMan,
                 WorkItem* wi)
{
    OS_ACQUIRE_LOCK(&ioMan->active_work_lock);
    wi->link = ioMan->active_work_items;
    ioMan->active_work_items = wi;
    OS_RELEASE_LOCK(&ioMan->active_work_lock);
}

static
void
DeregisterWorkItem(IOManagerState* ioMan,
                   WorkItem* wi)
{
    WorkItem *ptr, *prev;

    OS_ACQUIRE_LOCK(&ioMan->active_work_lock);
    for(prev=NULL,ptr=ioMan->active_work_items;ptr;prev=ptr,ptr=ptr->link) {
        if (wi->requestID == ptr->requestID) {
            if (prev==NULL) {
                ioMan->active_work_items = ptr->link;
            } else {
                prev->link = ptr->link;
            }
            OS_RELEASE_LOCK(&ioMan->active_work_lock);
            return;
        }
    }
    fprintf(stderr, "DeregisterWorkItem: unable to locate work item %d\n",
            wi->requestID);
    OS_RELEASE_LOCK(&ioMan->active_work_lock);
}


/*
 * Function: abandonWorkRequest()
 *
 * Signal that a work request isn't of interest. Called by the Scheduler
 * if a blocked Haskell thread has an exception thrown to it.
 *
 * Note: we're not aborting the system call that a worker might be blocked on
 * here, just disabling the propagation of its result once its finished. We
 * may have to go the whole hog here and switch to overlapped I/O so that we
 * can abort blocked system calls.
 */
void
abandonWorkRequest ( int reqID )
{
    WorkItem *ptr;
    OS_ACQUIRE_LOCK(&ioMan->active_work_lock);
    for(ptr=ioMan->active_work_items;ptr;ptr=ptr->link) {
        if (ptr->requestID == (unsigned int)reqID ) {
            ptr->abandonOp = 1;
            OS_RELEASE_LOCK(&ioMan->active_work_lock);
            return;
        }
    }
    /* Note: if the request ID isn't present, the worker will have
     * finished sometime since awaitRequests() last drained the completed
     * request table; i.e., not an error.
     */
    OS_RELEASE_LOCK(&ioMan->active_work_lock);
}

#endif