aboutsummaryrefslogtreecommitdiffstats
path: root/erts/emulator/sys/common/erl_poll.c
blob: c671ea99f4b5303832c3801f905c6a4be01d8415 (plain) (blame)
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
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
/*
 * %CopyrightBegin%
 *
 * Copyright Ericsson AB 2006-2016. All Rights Reserved.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 * %CopyrightEnd%
 */

/**
 * @description Poll interface suitable for ERTS
 *
 *		The interface is currently implemented using:
 *		- select
 *		- poll
 *              - /dev/poll
 *              - epoll with poll or select as fallback
 *              - kqueue with poll or select as fallback
 *
 *
 * @author Rickard Green
 * @author Lukas Larsson
 *
 * There are two major different implementations off IO polling in this
 * file. The concurrent and non-concurrent implementations.
 * When available epoll/kqueue are used to implement the concurrent
 * versions. poll, select and dev/poll use non-concurrent updates.
 *
 * Concurrent version:
 *   In the concurrent version erts_poll_control directly modifies
 *   the kernel pollset without waking the thread that is waiting
 *   on events. Also the ErtsPollResFd type is directly mapped to
 *   the native event type, so no extra copying is needed. Note that
 *   as no locking at all is done, fds can be triggered that have been
 *   removed from the pollset. The check_io layer has to deal with this.
 *
 * Non-concurrent version:
 *   In the non-concurrent version, the pollset has an internal representation
 *   of the pollset that is updated by erts_poll_control. When an fd is updated,
 *   its number is placed in the update request queue and then the waiting thread
 *   is woken in order to see the change. The internal data in the pollset is
 *   protected by a mutex that has to be taken by both the modifying and waiting
 *   thread at different times.
 *
 *   The non-concurrent pollset cannot have fd's closed in it while a thread is
 *   waiting on that fd. In order to fix this, when an ERTS_POLL_OP_DEL command
 *   is issued, the fd is marked as closing and the waiting thread is woken. The
 *   fd is then returned in the waiting threads results as ERTS_POLL_EV_NONE.
 *
 */

#ifdef HAVE_CONFIG_H
#  include "config.h"
#endif

#if defined(__DARWIN__) || defined(__APPLE__) && defined(__MACH__)
/* Setting _DARWIN_UNLIMITED_SELECT before including sys/select.h enables
 * the version of select() that does not place a limit on the fd_set.
 */
#  define _DARWIN_UNLIMITED_SELECT
#endif

#ifndef WANT_NONBLOCKING
#  define WANT_NONBLOCKING
#endif

#include "erl_poll.h"
#if ERTS_POLL_USE_KQUEUE
#  include <sys/types.h>
#  include <sys/event.h>
#  include <sys/time.h>
#endif
#if ERTS_POLL_USE_SELECT
#  ifdef SYS_SELECT_H
#    include <sys/select.h>
#  endif
#elif defined(_DARWIN_UNLIMITED_SELECT)
#  undef _DARWIN_UNLIMITED_SELECT
#endif
#ifdef NO_SYSCONF
#  if ERTS_POLL_USE_SELECT
#    include <sys/param.h>
#  else
#    include <limits.h>
#  endif
#endif
#include "erl_thr_progress.h"
#include "erl_driver.h"
#include "erl_alloc.h"
#include "erl_msacc.h"
#include "erl_misc_utils.h"

#if !defined(ERTS_POLL_USE_EPOLL) \
    && !defined(ERTS_POLL_USE_DEVPOLL)  \
    && !defined(ERTS_POLL_USE_POLL) \
    && !defined(ERTS_POLL_USE_SELECT)
#error "Missing implementation of erts_poll()"
#endif

#if 0
#define ERTS_POLL_DEBUG_PRINT 1

#define DEBUG_PRINT(FMT, PS, ...)                                       \
    do {                                                                \
        int myerrno = errno;                                            \
        erts_printf("%d: " FMT "\r\n", (PS)->id, ##__VA_ARGS__);        \
        errno = myerrno;                                                \
    } while(0)

/* Define to print info about modifications done to each fd */
#define DEBUG_PRINT_FD(FMT, PS, FD, ...) DEBUG_PRINT("%d: " FMT, PS, FD, ##__VA_ARGS__)
/* Define to print entry and exit from erts_poll_wait (can be very spammy) */
//#define DEBUG_PRINT_WAIT(FMT, PS, ...) DEBUG_PRINT(FMT, PS, ##__VA_ARGS__)

#else
#define ERTS_POLL_DEBUG_PRINT 0
#define DEBUG_PRINT(...)
#endif

#ifndef DEBUG_PRINT_FD
#define DEBUG_PRINT_FD(...)
#endif
#ifndef DEBUG_PRINT_WAIT
#define DEBUG_PRINT_WAIT(...)
#endif


#if defined(_DARWIN_UNLIMITED_SELECT) && ERTS_POLL_USE_SELECT
typedef struct {
    size_t sz;
    fd_set* ptr;
}ERTS_fd_set;
#  define ERTS_FD_CLR(fd, fds)	FD_CLR((fd), (fds)->ptr)
#  define ERTS_FD_SET(fd, fds)	FD_SET((fd), (fds)->ptr)
#  define ERTS_FD_ISSET(fd,fds) FD_ISSET((fd), (fds)->ptr)
#  define ERTS_FD_ZERO(fds)	memset((fds)->ptr, 0, (fds)->sz)
#  define ERTS_FD_SIZE(n)	((((n)+NFDBITS-1)/NFDBITS)*sizeof(fd_mask))

static void ERTS_FD_COPY(ERTS_fd_set *src, ERTS_fd_set *dst)
{
    if (dst->sz != src->sz) {
	dst->ptr = dst->ptr
	    ? erts_realloc(ERTS_ALC_T_SELECT_FDS, dst->ptr, src->sz)
	    : erts_alloc(ERTS_ALC_T_SELECT_FDS, src->sz);
	dst->sz = src->sz;
    }
    memcpy(dst->ptr, src->ptr, src->sz);
}

static ERTS_INLINE
int ERTS_SELECT(int nfds, ERTS_fd_set *readfds, ERTS_fd_set *writefds,
		ERTS_fd_set *exceptfds, struct timeval *timeout)
{
    ASSERT(!readfds || readfds->sz >= ERTS_FD_SIZE(nfds));
    ASSERT(!writefds || writefds->sz >= ERTS_FD_SIZE(nfds));
    ASSERT(!exceptfds);
    return select(nfds, 
		  (readfds ? readfds->ptr : NULL ),
		  (writefds ? writefds->ptr : NULL),
		  NULL,
		  timeout);
}

#else /* !_DARWIN_UNLIMITED_SELECT */
#  define ERTS_fd_set	fd_set
#  define ERTS_FD_CLR	FD_CLR
#  define ERTS_FD_ISSET FD_ISSET
#  define ERTS_FD_SET	FD_SET
#  define ERTS_FD_ZERO	FD_ZERO
#  define ERTS_FD_COPY(src,dst) (*(dst) = *(src))
#  define ERTS_SELECT	select
#endif

#define ERTS_POLL_IS_FALLBACK (ERTS_POLL_USE_POLL || ERTS_POLL_USE_SELECT) && ERTS_ENABLE_KERNEL_POLL

#define ERTS_POLL_USE_CONCURRENT_UPDATE (ERTS_POLL_USE_EPOLL || ERTS_POLL_USE_KQUEUE)

#if !ERTS_POLL_USE_CONCURRENT_UPDATE

#define ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(PS) \
  erts_atomic32_set_nob(&(PS)->have_update_requests, (erts_aint32_t) 1)
#define ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(PS) \
  erts_atomic32_set_nob(&(PS)->have_update_requests, (erts_aint32_t) 0)
#define ERTS_POLLSET_HAVE_UPDATE_REQUESTS(PS) \
  ((int) erts_atomic32_read_nob(&(PS)->have_update_requests))

#define ERTS_POLLSET_LOCK(PS) \
  erts_mtx_lock(&(PS)->mtx)
#define ERTS_POLLSET_UNLOCK(PS) \
  erts_mtx_unlock(&(PS)->mtx)

#else

#define ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(PS)
#define ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(PS)
#define ERTS_POLLSET_HAVE_UPDATE_REQUESTS(PS) 0

#define ERTS_POLLSET_LOCK(PS)
#define ERTS_POLLSET_UNLOCK(PS)

#endif

/*
 * --- Data types ------------------------------------------------------------
 */

#if !ERTS_POLL_USE_CONCURRENT_UPDATE

#define ERTS_POLLSET_UPDATE_REQ_BLOCK_SIZE 128

typedef struct ErtsPollSetUpdateRequestsBlock_ ErtsPollSetUpdateRequestsBlock;
struct ErtsPollSetUpdateRequestsBlock_ {
    ErtsPollSetUpdateRequestsBlock *next;
    int len;
    int fds[ERTS_POLLSET_UPDATE_REQ_BLOCK_SIZE];
};

#  define ERTS_POLL_FD_FLG_INURQ	(((unsigned short) 1) << 0)
#  define ERTS_POLL_FD_FLG_RST		(((unsigned short) 1) << 1)

typedef struct {
#if ERTS_POLL_USE_POLL
    int pix;
#endif

    ErtsPollEvents used_events;
    ErtsPollEvents events;
    unsigned short flags;

} ErtsFdStatus;

#endif

/*
 * This struct is not really exported, but it's nice to
 * get unique names in debugger for kp/nkp
 */
struct ERTS_POLL_EXPORT(erts_pollset) {
    int id;
    int internal_fd_limit;
    erts_atomic_t no_of_user_fds;

#if ERTS_POLL_USE_KERNEL_POLL
    int kp_fd;
#endif /* ERTS_POLL_USE_KERNEL_POLL */

#if ERTS_POLL_USE_POLL
    int next_poll_fds_ix;
    int no_poll_fds;
    int poll_fds_len;
    struct pollfd *poll_fds;
#elif ERTS_POLL_USE_SELECT
    int next_sel_fd;
    int max_fd;
    ERTS_fd_set input_fds;
    ERTS_fd_set res_input_fds;
    ERTS_fd_set output_fds;
    ERTS_fd_set res_output_fds;
#elif ERTS_POLL_USE_DEVPOLL
    struct pollfd *poll_fds;
    int poll_fds_ix;
#endif

#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    ErtsFdStatus *fds_status;
    int fds_status_len;
    ErtsPollSetUpdateRequestsBlock update_requests;
    ErtsPollSetUpdateRequestsBlock *curr_upd_req_block;
    erts_atomic32_t have_update_requests;
    erts_mtx_t mtx;
    int wake_fds[2];
    erts_atomic32_t wakeup_state;
#endif
};

void erts_silence_warn_unused_result(long unused);
static void fatal_error(char *format, ...);

static int max_fds = -1;

#if ERTS_POLL_USE_POLL

#if !ERTS_POLL_IS_FALLBACK
static ERTS_INLINE short ev2pollev(ErtsPollEvents ev)
{
    return ERTS_POLL_EV_E2N(ev);
}

static ERTS_INLINE ErtsPollEvents pollev2ev(short ev)
{
    return ERTS_POLL_EV_N2E(ev);
}

#else /* ERTS_POLL_IS_FALLBACK */

static ERTS_INLINE short
ev2pollev(ErtsPollEvents ev)
{
    short res_ev = (short) 0;
    if (ev & ERTS_POLL_EV_IN)
       res_ev |= ERTS_POLL_EV_NKP_IN;
    if (ev & ERTS_POLL_EV_OUT)
       res_ev |= ERTS_POLL_EV_NKP_OUT;
    return res_ev;
}

static ERTS_INLINE ErtsPollEvents
pollev2ev(short ev)
{
    ErtsPollEvents res_ev = (ErtsPollEvents) 0;
    if (ev & ERTS_POLL_EV_NKP_IN)
       res_ev |= ERTS_POLL_EV_IN;
    if (ev & ERTS_POLL_EV_NKP_OUT)
       res_ev |= ERTS_POLL_EV_OUT;
    if (ev & ERTS_POLL_EV_NKP_ERR)
       res_ev |= ERTS_POLL_EV_ERR;
    if (ev & ERTS_POLL_EV_NKP_NVAL)
       res_ev |= ERTS_POLL_EV_NVAL;
   return res_ev;
}

#endif /* !ERTS_POLL_IS_FALLBACK */

#endif /* ERTS_POLL_USE_POLL */


#ifdef HARD_DEBUG
static void check_poll_result(ErtsPollResFd pr[], int len);
#endif /* HARD_DEBUG */
#if ERTS_POLL_USE_DEVPOLL && defined(DEBUG)
static void check_poll_status(ErtsPollSet *ps);
#endif /* ERTS_POLL_USE_DEVPOLL && DEBUG */
static void print_misc_debug_info(void);
#if ERTS_POLL_USE_EPOLL
uint32_t epoll_events(int kp_fd, int fd);
#endif


#define ERTS_POLL_NOT_WOKEN	0
#define ERTS_POLL_WOKEN		-1
#define ERTS_POLL_WOKEN_INTR	1

#if !ERTS_POLL_USE_CONCURRENT_UPDATE
static ERTS_INLINE void
reset_wakeup_state(ErtsPollSet *ps)
{
    erts_atomic32_set_mb(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN);
}
#endif

static ERTS_INLINE int
is_woken(ErtsPollSet *ps)
{
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    return erts_atomic32_read_acqb(&ps->wakeup_state) != ERTS_POLL_NOT_WOKEN;
#else
    return 0;
#endif
}

static ERTS_INLINE int
is_interrupted_reset(ErtsPollSet *ps)
{
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    return (erts_atomic32_xchg_acqb(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN)
	    == ERTS_POLL_WOKEN_INTR);
#else
    return 0;
#endif
}

static ERTS_INLINE void
woke_up(ErtsPollSet *ps)
{
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    erts_aint32_t wakeup_state = erts_atomic32_read_acqb(&ps->wakeup_state);
    if (wakeup_state == ERTS_POLL_NOT_WOKEN)
	(void) erts_atomic32_cmpxchg_nob(&ps->wakeup_state,
					 ERTS_POLL_WOKEN,
					 ERTS_POLL_NOT_WOKEN);
    ASSERT(erts_atomic32_read_nob(&ps->wakeup_state) != ERTS_POLL_NOT_WOKEN);
#endif
}

/*
 * --- Wakeup pipe -----------------------------------------------------------
 */

#if !ERTS_POLL_USE_CONCURRENT_UPDATE

static ERTS_INLINE void
wake_poller(ErtsPollSet *ps, int interrupted)
{
    int wake;
    erts_aint32_t wakeup_state;
    if (!interrupted)
        wakeup_state = erts_atomic32_cmpxchg_relb(&ps->wakeup_state,
                                                  ERTS_POLL_WOKEN,
                                                  ERTS_POLL_NOT_WOKEN);
    else
        wakeup_state = erts_atomic32_xchg_relb(&ps->wakeup_state,
                                               ERTS_POLL_WOKEN_INTR);
    wake = wakeup_state == ERTS_POLL_NOT_WOKEN;

    if (wake) {
	ssize_t res;
	if (ps->wake_fds[1] < 0)
	    return; /* Not initialized yet */
	do {
	    /* write() is async-signal safe (according to posix) */
	    res = write(ps->wake_fds[1], "!", 1);
	} while (res < 0 && errno == EINTR);
	if (res <= 0 && errno != ERRNO_BLOCK) {
            fatal_error("%s:%d:wake_poller(): "
                        "Failed to write to wakeup pipe fd=%d: "
                        "%s (%d)\n",
                        __FILE__, __LINE__,
                        ps->wake_fds[1],
                        erl_errno_id(errno), errno);
	}
    }
}

static ERTS_INLINE void
cleanup_wakeup_pipe(ErtsPollSet *ps)
{
    int intr = 0;
    int fd = ps->wake_fds[0];
    int res;
    do {
	char buf[32];
	res = read(fd, buf, sizeof(buf));
	if (res > 0)
	    intr = 1;
    } while (res > 0 || (res < 0 && errno == EINTR));
    if (res < 0 && errno != ERRNO_BLOCK) {
	fatal_error("%s:%d:cleanup_wakeup_pipe(): "
		    "Failed to read on wakeup pipe fd=%d: "
		    "%s (%d)\n",
		    __FILE__, __LINE__,
		    fd,
		    erl_errno_id(errno), errno);
    }
    if (intr)
	erts_atomic32_set_nob(&ps->wakeup_state, ERTS_POLL_WOKEN_INTR);
}

static void
create_wakeup_pipe(ErtsPollSet *ps)
{
    int do_wake = 0;
    int wake_fds[2];
    ps->wake_fds[0] = -1;
    ps->wake_fds[1] = -1;
    if (pipe(wake_fds) < 0) {
	fatal_error("%s:%d:create_wakeup_pipe(): "
		    "Failed to create pipe: %s (%d)\n",
		    __FILE__,
		    __LINE__,
		    erl_errno_id(errno),
		    errno);
    }
    SET_NONBLOCKING(wake_fds[0]);
    SET_NONBLOCKING(wake_fds[1]);

    DEBUG_PRINT("wakeup fds = {%d, %d}", ps, wake_fds[0], wake_fds[1]);

    ERTS_POLL_EXPORT(erts_poll_control)(ps,
					wake_fds[0],
                                        ERTS_POLL_OP_ADD,
					ERTS_POLL_EV_IN,
                                        &do_wake);
    if (ps->internal_fd_limit <= wake_fds[1])
	ps->internal_fd_limit = wake_fds[1] + 1;
    if (ps->internal_fd_limit <= wake_fds[0])
	ps->internal_fd_limit = wake_fds[0] + 1;
    ps->wake_fds[0] = wake_fds[0];
    ps->wake_fds[1] = wake_fds[1];
}


/*
 * --- Poll set update requests ----------------------------------------------
 */

static ERTS_INLINE void
enqueue_update_request(ErtsPollSet *ps, int fd)
{
    ErtsPollSetUpdateRequestsBlock *urqbp;

    ASSERT(fd < ps->fds_status_len);

    if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INURQ)
	return;

    if (ps->update_requests.len == 0)
	ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(ps);

    urqbp = ps->curr_upd_req_block;

    if (urqbp->len == ERTS_POLLSET_UPDATE_REQ_BLOCK_SIZE) {
	urqbp = erts_alloc(ERTS_ALC_T_POLLSET_UPDREQ,
			   sizeof(ErtsPollSetUpdateRequestsBlock));
	urqbp->next = ps->curr_upd_req_block;
	urqbp->len = 0;
	ps->curr_upd_req_block = urqbp;
    }

    ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_INURQ;
    urqbp->fds[urqbp->len++] = fd;
}

static ERTS_INLINE void
free_update_requests_block(ErtsPollSet *ps,
			   ErtsPollSetUpdateRequestsBlock *urqbp)
{
    if (urqbp != &ps->update_requests)
	erts_free(ERTS_ALC_T_POLLSET_UPDREQ, (void *) urqbp);
    else {
	urqbp->len = 0;
    }
}

#endif /* !ERTS_POLL_USE_CONCURRENT_UPDATE */

/*
 * --- Growing poll set structures -------------------------------------------
 */

#if !ERTS_NO_KERNEL_POLL_VERSION || !ERTS_ENABLE_KERNEL_POLL
/* only one shared implementation */

#define ERTS_FD_TABLE_MIN_LENGTH	1024
#define ERTS_FD_TABLE_EXP_THRESHOLD	(2048*1024)

int erts_poll_new_table_len(int old_len, int need_len)
{
    int new_len;

    ASSERT(need_len > old_len);
    if (need_len < ERTS_FD_TABLE_MIN_LENGTH) {
	new_len = ERTS_FD_TABLE_MIN_LENGTH;
    }
    else {
        new_len = old_len;
        do {
            if (new_len < ERTS_FD_TABLE_EXP_THRESHOLD)
                new_len *= 2;
            else
                new_len += ERTS_FD_TABLE_EXP_THRESHOLD;

        } while (new_len < need_len);
    }
    ASSERT(new_len >= need_len);
    return new_len;
}
#endif

#if ERTS_POLL_USE_POLL
static void
grow_poll_fds(ErtsPollSet *ps, int min_ix)
{
    int i;
    int new_len = erts_poll_new_table_len(ps->poll_fds_len, min_ix + 1);
    if (new_len > max_fds)
	new_len = max_fds;
    ps->poll_fds = (ps->poll_fds_len
		    ? erts_realloc(ERTS_ALC_T_POLL_FDS,
				   ps->poll_fds,
				   sizeof(struct pollfd)*new_len)
		    : erts_alloc(ERTS_ALC_T_POLL_FDS,
				 sizeof(struct pollfd)*new_len));
    for (i = ps->poll_fds_len; i < new_len; i++) {
	ps->poll_fds[i].fd = -1;
	ps->poll_fds[i].events = (short) 0;
	ps->poll_fds[i].revents = (short) 0;
    }
    ps->poll_fds_len = new_len;
}
#endif

#ifdef _DARWIN_UNLIMITED_SELECT
static void
grow_select_fds(int fd, ERTS_fd_set* fds)
{
    int new_len = erts_poll_new_table_len(fds->sz, fd + 1);
    if (new_len > max_fds)
	new_len = max_fds;
    new_len = ERTS_FD_SIZE(new_len);
    fds->ptr = fds->sz
	? erts_realloc(ERTS_ALC_T_SELECT_FDS, fds->ptr, new_len)
	: erts_alloc(ERTS_ALC_T_SELECT_FDS, new_len);
    memset((char*)fds->ptr + fds->sz, 0, new_len - fds->sz);
    fds->sz = new_len;
}
static ERTS_INLINE void
ensure_select_fds(int fd, ERTS_fd_set* in, ERTS_fd_set* out)
{
    ASSERT(in->sz == out->sz);
    if (ERTS_FD_SIZE(fd+1) > in->sz) {
	grow_select_fds(fd, in);
	grow_select_fds(fd, out);
    }
}
#else
#  define ensure_select_fds(fd, in, out) do {} while(0)
#endif /* _DARWIN_UNLIMITED_SELECT */

#if !ERTS_POLL_USE_CONCURRENT_UPDATE
static void
grow_fds_status(ErtsPollSet *ps, int min_fd)
{
    int i;
    int new_len = erts_poll_new_table_len(ps->fds_status_len, min_fd + 1);
    ASSERT(min_fd < max_fds);
    if (new_len > max_fds)
	new_len = max_fds;
    ps->fds_status = (ps->fds_status_len
		      ? erts_realloc(ERTS_ALC_T_FD_STATUS,
				     ps->fds_status,
				     sizeof(ErtsFdStatus)*new_len)
		      : erts_alloc(ERTS_ALC_T_FD_STATUS,
				   sizeof(ErtsFdStatus)*new_len));
    for (i = ps->fds_status_len; i < new_len; i++) {
#if ERTS_POLL_USE_POLL
	ps->fds_status[i].pix = -1;
#endif
	ps->fds_status[i].used_events = (ErtsPollEvents) 0;
	ps->fds_status[i].events = (ErtsPollEvents) 0;
	ps->fds_status[i].flags = (unsigned short) 0;
    }
    ps->fds_status_len = new_len;
}
#endif

/*
 * --- Selecting fd to poll on -----------------------------------------------
 */

#if ERTS_POLL_USE_EPOLL
static int
update_pollset(ErtsPollSet *ps, int fd, ErtsPollOp op, ErtsPollEvents events)
{
    int res;
    int epoll_op = EPOLL_CTL_MOD;
    struct epoll_event epe_templ;
    struct epoll_event epe;

    epe_templ.events = ERTS_POLL_EV_E2N(events) | EPOLLONESHOT;
    epe_templ.data.fd = fd;

#ifdef VALGRIND
    /* Silence invalid valgrind warning ... */
    memset((void *) &epe.data, 0, sizeof(epoll_data_t));
#endif

    switch (op) {
    case ERTS_POLL_OP_DEL:
	/* A note on EPOLL_CTL_DEL: linux kernel versions before 2.6.9
	   need a non-NULL event pointer even though it is ignored... */
        epoll_op = EPOLL_CTL_DEL;
        epe_templ.events = 0;
	erts_atomic_dec_nob(&ps->no_of_user_fds);
        break;
    case ERTS_POLL_OP_ADD:
        epoll_op = EPOLL_CTL_ADD;
	erts_atomic_inc_nob(&ps->no_of_user_fds);
        break;
    case ERTS_POLL_OP_MOD:
        epoll_op = EPOLL_CTL_MOD;
        break;
    default:
        ASSERT(0);
        break;
    }

    do {
	/* We init 'epe' every time since epoll_ctl() may modify it
	   (not declared const and not documented as const). */
	epe.events = epe_templ.events;
	epe.data.fd = epe_templ.data.fd;
	res = epoll_ctl(ps->kp_fd, epoll_op, fd, &epe);
    } while (res != 0 && errno == EINTR);

#if ERTS_POLL_DEBUG_PRINT
    {
	int saved_errno = errno;
	DEBUG_PRINT_FD("%s = epoll_ctl(%d, %s, %d, {0x%x, %d})",
                       ps, fd,
                       res == 0 ? "0" : erl_errno_id(errno),
                       ps->kp_fd,
                       (epoll_op == EPOLL_CTL_ADD
                        ? "EPOLL_CTL_ADD"
                        : (epoll_op == EPOLL_CTL_MOD
                           ? "EPOLL_CTL_MOD"
                           : (epoll_op == EPOLL_CTL_DEL
                              ? "EPOLL_CTL_DEL"
                              : "UNKNOWN"))),
                       fd,
                       epe_templ.events,
                       fd);
	errno = saved_errno;
    }
#endif
    if (res != 0) {
	switch (op) {
	case ERTS_POLL_OP_MOD:
	    epe.events = 0;
	    do {
		/* We init 'epe' every time since epoll_ctl() may modify it
		   (not declared const and not documented as const). */
		epe.events = 0;
		epe.data.fd = fd;
		res = epoll_ctl(ps->kp_fd, EPOLL_CTL_DEL, fd, &epe);
	    } while (res != 0 && errno == EINTR);
	/* Fall through ... */
	case ERTS_POLL_OP_ADD: {
	    erts_atomic_dec_nob(&ps->no_of_user_fds);
            res = ERTS_POLL_EV_NVAL;
	    break;
	}
	case ERTS_POLL_OP_DEL: {
	    /*
	     * Since we use a lazy update approach EPOLL_CTL_DEL will
	     * frequently fail. This since epoll automatically removes
	     * a filedescriptor that is closed from the poll set.
	     */
	    res = 0;
	    break;
	}
	default:
	    fatal_error("%s:%d:update_pollset(): Internal error\n",
			__FILE__, __LINE__);
	    break;
	}
    } else {
        res = events;
    }
    return res;
}

#endif /* ERTS_POLL_USE_EPOLL */

#if ERTS_POLL_USE_KQUEUE

/* Some versions of the EV_SET macro used kevp multiple times,
   so we define out own version that make sure that it is safe
   to do kevp++ in the argument list. */
#define ERTS_EV_SET(kevp, a, b, c, f) do {              \
        struct kevent *kevp_ = kevp;                    \
        EV_SET(kevp_, a, b, c, 0, 0, f);                \
    } while(0)

static int
update_pollset(ErtsPollSet *ps, int fd, ErtsPollOp op, ErtsPollEvents events)
{
    int res = 0, len = 0;
    struct kevent evts[2];
    struct timespec ts = {0, 0};

#ifdef EV_DISPATCH
    /* If we have EV_DISPATCH we use it. The kevent descriptions for both
       read and write are added on OP_ADD and removed on OP_DEL. And then
       after than only EV_ENABLE|EV_DISPATCH are used.

       It could be possible to not modify the pollset when disabling and/or
       deleting events, but that may cause the poll threads to be awoken
       a lot more than they should so we take the cost here instead of
       in the poll thread.

       Note: We need to have EV_DISPATCH both when the event is enabled and
       disabled, as otherwise the event may be triggered twice on each re-arm.
       Not sure if this is intended or not (can't find anything about it in the
       man page), but it seems to be the way it works...
    */

    if (op == ERTS_POLL_OP_DEL) {
        erts_atomic_dec_nob(&ps->no_of_user_fds);
        /* We could probably skip this delete, do we want to? */
        ERTS_EV_SET(&evts[len++], fd, EVFILT_READ, EV_DELETE, (void *) 0);
        ERTS_EV_SET(&evts[len++], fd, EVFILT_WRITE, EV_DELETE, (void *) 0);
    } else if (op == ERTS_POLL_OP_ADD) {
        uint32_t flags;
        erts_atomic_inc_nob(&ps->no_of_user_fds);

        flags = EV_ADD|EV_DISPATCH;
        flags |= ((events & ERTS_POLL_EV_IN) ? 0 : EV_DISABLE);
        ERTS_EV_SET(&evts[len++], fd, EVFILT_READ, flags, (void *) ERTS_POLL_EV_IN);

        flags = EV_ADD|EV_DISPATCH;
        flags |= ((events & ERTS_POLL_EV_OUT) ? 0 : EV_DISABLE);
        ERTS_EV_SET(&evts[len++], fd, EVFILT_WRITE, flags, (void *) ERTS_POLL_EV_OUT);
    } else {
        uint32_t flags;
        ASSERT(op == ERTS_POLL_OP_MOD);

        flags = EV_DISPATCH;
        flags |= (events & ERTS_POLL_EV_IN) ? EV_ENABLE : EV_DISABLE;
        ERTS_EV_SET(&evts[len++], fd, EVFILT_READ, flags, (void *) ERTS_POLL_EV_IN);

        flags = EV_DISPATCH;
        flags |= (events & ERTS_POLL_EV_OUT) ? EV_ENABLE : EV_DISABLE;
        ERTS_EV_SET(&evts[len++], fd, EVFILT_WRITE, flags, (void *) ERTS_POLL_EV_OUT);
    }
#else
    uint32_t flags = EV_ADD|EV_ONESHOT;

    if (op == ERTS_POLL_OP_DEL) {
        erts_atomic_dec_nob(&ps->no_of_user_fds);
        /* We don't do anything when a delete is issued. The fds will be removed
           when they are triggered, or when they are closed. */
        events = 0;
    } else if (op == ERTS_POLL_OP_ADD) {
        erts_atomic_inc_nob(&ps->no_of_user_fds);
    }

    if (events & ERTS_POLL_EV_IN) {
        ERTS_EV_SET(&evts[len++], fd, EVFILT_READ, flags, (void *) ERTS_POLL_EV_IN);
    }
    if (events & ERTS_POLL_EV_OUT) {
        ERTS_EV_SET(&evts[len++], fd, EVFILT_WRITE, flags, (void *) ERTS_POLL_EV_OUT);
    }

#endif
    if (len)
        do {
            res = kevent(ps->kp_fd, evts, len, NULL, 0, &ts);
        } while (res < 0 && errno == EINTR);
#if ERTS_POLL_DEBUG_PRINT
    {
        int saved_errno = errno, i;
        char keventb[255], *keventbp = keventb;
        if (res < 0)
            keventbp += sprintf(keventbp,"%s = ",erl_errno_id(saved_errno));
        else
            keventbp += sprintf(keventbp,"%d = ",res);
        keventbp += sprintf(keventbp, "kevent(%d, {",ps->kp_fd);
        for (i = 0; i < len; i++) {
            const char *flags = "UNKNOWN";
            if (evts[i].flags == EV_DELETE) flags = "EV_DELETE";
            if (evts[i].flags == (EV_ADD|EV_ONESHOT)) flags = "EV_ADD|EV_ONESHOT";
#ifdef EV_DISPATCH
            if (evts[i].flags == (EV_ADD|EV_DISPATCH)) flags = "EV_ADD|EV_DISPATCH";
            if (evts[i].flags == (EV_ADD|EV_DISABLE)) flags = "EV_ADD|EV_DISABLE";
            if (evts[i].flags == (EV_ENABLE|EV_DISPATCH)) flags = "EV_ENABLE|EV_DISPATCH";
            if (evts[i].flags == EV_DISABLE) flags = "EV_DISABLE";
            if (evts[i].flags == (EV_DISABLE|EV_DISPATCH)) flags = "EV_DISABLE|EV_DISABLE";
#endif

            keventbp += sprintf(keventbp, "%s{%lu, %s, %s}",i > 0 ? ", " : "",
                                evts[i].ident,
                                (evts[i].filter == EVFILT_READ
                                 ? "EVFILT_READ"
                                 : (evts[i].filter == EVFILT_WRITE
                                    ? "EVFILT_WRITE"
                                    : "UNKNOWN")), flags);
        }
        keventbp += sprintf(keventbp, "}, %d)", len);
        DEBUG_PRINT_FD("%s", ps, fd, keventb);
        errno = saved_errno;
    }
#endif
    if (res < 0) {
        if (op != ERTS_POLL_OP_DEL) {
#ifdef EV_RECEIPT
            struct kevent receipt_evts[2];
            len = 0;
            ERTS_EV_SET(&evts[len++], fd, EVFILT_WRITE, EV_DELETE|EV_RECEIPT, (void *) 0);
            ERTS_EV_SET(&evts[len++], fd, EVFILT_READ, EV_DELETE|EV_RECEIPT, (void *) 0);
            do {
                res = kevent(ps->kp_fd, evts, len, receipt_evts, 2, &ts);
            } while (res < 0 && errno == EINTR);
#else
            ERTS_EV_SET(&evts[0], fd, EVFILT_WRITE, EV_DELETE, (void *) 0);
            do {
                res = kevent(ps->kp_fd, evts, 1, NULL, 0, &ts);
            } while (res < 0 && errno == EINTR);
            ERTS_EV_SET(&evts[0], fd, EVFILT_READ, EV_DELETE, (void *) 0);
            do {
                res = kevent(ps->kp_fd, evts, 1, NULL, 0, &ts);
            } while (res < 0 && errno == EINTR);
#endif
            if (op == ERTS_POLL_OP_ADD)
                erts_atomic_dec_nob(&ps->no_of_user_fds);
            return ERTS_POLL_EV_NVAL;
        }
        return 0;
    }
    return events;
}

#endif /* ERTS_POLL_USE_KQUEUE */

#if !ERTS_POLL_USE_CONCURRENT_UPDATE

static ERTS_INLINE void
init_batch_update(ErtsPollSet *ps, int len)
{
#if ERTS_POLL_USE_DEVPOLL
    ASSERT(ps->poll_fds == NULL);
    ps->poll_fds = erts_alloc(ERTS_ALC_T_TMP, sizeof(struct pollfd) * len);
    ps->poll_fds_ix = 0;
#endif
}

static ERTS_INLINE void
write_batch_update(ErtsPollSet *ps)
{
#if ERTS_POLL_USE_DEVPOLL
    ssize_t wres;
    char *buf = (char *) ps->poll_fds;
    size_t buf_size = sizeof(struct pollfd)*ps->poll_fds_ix;

    while (1) {
	wres = write(ps->kp_fd, (void *) buf, buf_size);
	if (wres < 0) {
	    if (errno == EINTR)
		continue;
	    fatal_error("%s:%d:write_batch_buf(): "
			"Failed to write to /dev/poll: "
			"%s (%d)\n",
			__FILE__, __LINE__,
			erl_errno_id(errno), errno);
	}
#if ERTS_POLL_DEBUG_PRINT
        {
            int saved_errno = errno, i;
            char devpollb[2048], *devpollbp = devpollb;
            devpollbp += sprintf(devpollbp, "%d = devpoll(%d, {", wres, ps->kp_fd);
            for (i = 0; i < wres / sizeof(struct pollfd); i++) {
                if (devpollbp == devpollb)
                    devpollbp += sprintf(devpollbp, "%d = devpoll(%d, {", wres, ps->kp_fd);
                devpollbp += sprintf(devpollbp, "%s{fd = %d, events = %s}",
                                     i > 0 ? ", " : "",
                                     ps->poll_fds[i].fd,
                                     ev2str(ps->poll_fds[i].events));
                if (devpollbp - devpollb > 512) {
                    devpollbp += sprintf(devpollbp, "}, %d)", ps->poll_fds_ix);
                    DEBUG_PRINT("%s", ps, devpollb);
                    devpollbp = devpollb;
                }
            }
            devpollbp += sprintf(devpollbp, "}, %d)", ps->poll_fds_ix);
            DEBUG_PRINT("%s", ps, devpollb);
            errno = saved_errno;
        }
#endif

	buf_size -= wres;
	if (buf_size <= 0)
	    break;
	buf += wres;
    }

    if (buf_size < 0) {
	fatal_error("%s:%d:write_devpoll_buf(): Internal error\n",
		    __FILE__, __LINE__);
    }
    erts_free(ERTS_ALC_T_TMP, ps->poll_fds);
    ps->poll_fds = NULL;
#endif
}

static ERTS_INLINE int
need_update(ErtsPollSet *ps, int fd, int *resetp)
{
    int reset;
    ASSERT(fd < ps->fds_status_len);

    reset = (int) (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_RST);
    ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_RST;

    *resetp = reset;

    if (reset || ps->fds_status[fd].used_events != ps->fds_status[fd].events)
	return 1;

    return 0;
}

static int update_pollset(ErtsPollSet *ps, ErtsPollResFd pr[], int fd)
{
    int res = 0, reset = 0;
    ErtsPollEvents events = ps->fds_status[fd].events;
    ASSERT(fd < ps->fds_status_len);

    if (!need_update(ps, fd, &reset))
	return res;

#if ERTS_POLL_USE_POLL	/* --- poll -------------------------------- */
    if (!events) {
	int pix = ps->fds_status[fd].pix;
	int last_pix;

        if (reset) {
            /* When a fd has been reset, we tell the caller of erts_poll_wait
               this by setting the fd as ERTS_POLL_EV_NONE */
            ERTS_POLL_RES_SET_FD(&pr[res], fd);
            ERTS_POLL_RES_SET_EVTS(&pr[res], ERTS_POLL_EV_NONE);
            DEBUG_PRINT_FD("trig %s (poll)", ps, fd, ev2str(ERTS_POLL_EV_NONE));
            res++;
        }

	if (pix < 0) {
	    return res;
	}
	erts_atomic_dec_nob(&ps->no_of_user_fds);
	last_pix = --ps->no_poll_fds;
	if (pix != last_pix) {
	/* Move last pix to this pix */
	    ps->poll_fds[pix].fd = ps->poll_fds[last_pix].fd;
	    ps->poll_fds[pix].events = ps->poll_fds[last_pix].events;
	    ps->poll_fds[pix].revents = ps->poll_fds[last_pix].revents;
	    ps->fds_status[ps->poll_fds[pix].fd].pix = pix;
	}
	/* Clear last pix */
	ps->poll_fds[last_pix].fd = -1;
	ps->poll_fds[last_pix].events = (short) 0;
	ps->poll_fds[last_pix].revents = (short) 0;
	/* Clear this fd status */
	ps->fds_status[fd].pix = -1;
	ps->fds_status[fd].used_events = (ErtsPollEvents) 0;

    }
    else {
	int pix = ps->fds_status[fd].pix;
	if (pix < 0) {
	    erts_atomic_inc_nob(&ps->no_of_user_fds);
	    ps->fds_status[fd].pix = pix = ps->no_poll_fds++;
	    if (pix >= ps->poll_fds_len)
		grow_poll_fds(ps, pix);
	    ps->poll_fds[pix].fd = fd;
	    ps->fds_status[fd].pix = pix;
	}

	/* Events to be used in next poll */
	ps->poll_fds[pix].events = ev2pollev(events);
	if (ps->poll_fds[pix].revents) {
	    /* Remove result events that we should not poll for anymore */
	    ps->poll_fds[pix].revents
		&= ev2pollev(~(~ps->fds_status[fd].used_events
			       & events));
	}
	/* Save events to be used in next poll */
	ps->fds_status[fd].used_events = events;
    }
    return res;
#elif ERTS_POLL_USE_SELECT	/* --- select ------------------------------ */
    if (!events) {

        if (reset) {
            /* When a fd has been reset, we tell the caller of erts_poll_wait
               this by setting the fd as ERTS_POLL_EV_NONE */
            ERTS_POLL_RES_SET_FD(&pr[res], fd);
            ERTS_POLL_RES_SET_EVTS(&pr[res], ERTS_POLL_EV_NONE);
            DEBUG_PRINT_FD("trig %s (select)", ps, fd, ev2str(ERTS_POLL_EV_NONE));
            res++;
        }

        ERTS_FD_CLR(fd, &ps->input_fds);
        ERTS_FD_CLR(fd, &ps->output_fds);

        if (ps->fds_status[fd].used_events) {
            erts_atomic_dec_nob(&ps->no_of_user_fds);
            ps->fds_status[fd].used_events = (ErtsPollEvents) 0;
        }

        if (fd == ps->max_fd) {
            int max = ps->max_fd;
            for (max = ps->max_fd; max >= 0; max--)
                if (ps->fds_status[max].used_events)
                    break;
            ps->max_fd = max;
        }

    } else {

	ensure_select_fds(fd, &ps->input_fds, &ps->output_fds);

        if (!ps->fds_status[fd].used_events)
            erts_atomic_inc_nob(&ps->no_of_user_fds);

        if (events & ERTS_POLL_EV_IN)
            ERTS_FD_SET(fd, &ps->input_fds);
        else
            ERTS_FD_CLR(fd, &ps->input_fds);

        if (events & ERTS_POLL_EV_OUT)
            ERTS_FD_SET(fd, &ps->output_fds);
        else
            ERTS_FD_CLR(fd, &ps->output_fds);

	ps->fds_status[fd].used_events = events;

        if (fd > ps->max_fd)
            ps->max_fd = fd;
    }

    return res;
#elif ERTS_POLL_USE_DEVPOLL

    if (!events) {

        if (reset) {
            /* When a fd has been reset, we tell the caller of erts_poll_wait
               this by setting the fd as ERTS_POLL_EV_NONE */
            ERTS_POLL_RES_SET_FD(&pr[res], fd);
            ERTS_POLL_RES_SET_EVTS(&pr[res], ERTS_POLL_EV_NONE);
            DEBUG_PRINT_FD("trig %s (devpoll)", ps, fd, ev2str(ERTS_POLL_EV_NONE));
            res++;
        }

        ps->poll_fds[ps->poll_fds_ix].fd = fd;
        ps->poll_fds[ps->poll_fds_ix].revents = 0;
        ps->poll_fds[ps->poll_fds_ix++].events = POLLREMOVE;

        if (ps->fds_status[fd].used_events) {
            erts_atomic_dec_nob(&ps->no_of_user_fds);
            ps->fds_status[fd].used_events = 0;
        }

    } else {
        if (!ps->fds_status[fd].used_events) {
            erts_atomic_inc_nob(&ps->no_of_user_fds);
        }
        ps->poll_fds[ps->poll_fds_ix].fd = fd;
        ps->poll_fds[ps->poll_fds_ix].revents = 0;
        ps->poll_fds[ps->poll_fds_ix++].events = ERTS_POLL_EV_E2N(events);
        ps->fds_status[fd].used_events = ps->fds_status[fd].events;
    }

    return res;
#endif
}

static int
handle_update_requests(ErtsPollSet *ps, ErtsPollResFd pr[], int no_fds)
{
    int res = 0;
    ErtsPollSetUpdateRequestsBlock *urqbp = ps->curr_upd_req_block;

    while (urqbp) {
	ErtsPollSetUpdateRequestsBlock *free_urqbp = urqbp;
	int i;
	int len = urqbp->len;

        init_batch_update(ps, len);

	for (i = 0; i < len; i++) {
	    int fd = urqbp->fds[i];
	    ASSERT(fd < ps->fds_status_len);
            ASSERT(ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INURQ);

            /* We have run out of PollResFd slots to put results in,
               so we yield here and return later for more. */
            if (res == no_fds && pr != NULL) {
                memmove(urqbp->fds, urqbp->fds+i, sizeof(int) * (len - i));
                urqbp->len -= i;
                ps->curr_upd_req_block = urqbp;
                write_batch_update(ps);
                return res;
            }

            if (ps->fds_status[fd].flags & ERTS_POLL_FD_FLG_INURQ) {
                ps->fds_status[fd].flags &= ~ERTS_POLL_FD_FLG_INURQ;
                res += update_pollset(ps, pr + res, fd);
            }
	}

	free_urqbp = urqbp;
	urqbp = urqbp->next;

	free_update_requests_block(ps, free_urqbp);

        write_batch_update(ps);

    }

    ps->curr_upd_req_block = &ps->update_requests;

#if ERTS_POLL_USE_DEVPOLL && defined(HARD_DEBUG)
    check_poll_status(ps);
#endif

    ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(ps);
    return res;
}

#endif /* !ERTS_POLL_USE_CONCURRENT_UPDATE */

static ERTS_INLINE ErtsPollEvents
poll_control(ErtsPollSet *ps, int fd, ErtsPollOp op,
             ErtsPollEvents events, int *do_wake)
{
    ErtsPollEvents new_events;

    if (fd < ps->internal_fd_limit || fd >= max_fds) {
	if (fd < 0 || fd >= max_fds) {
	    new_events = ERTS_POLL_EV_ERR;
	    goto done;
	}
#if ERTS_POLL_USE_KERNEL_POLL
	if (fd == ps->kp_fd) {
	    new_events = ERTS_POLL_EV_NVAL;
	    goto done;
	}
#endif
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
	if (fd == ps->wake_fds[0] || fd == ps->wake_fds[1]) {
	    new_events = ERTS_POLL_EV_NVAL;
	    goto done;
	}
#endif
    }

#if ERTS_POLL_USE_CONCURRENT_UPDATE

    new_events = update_pollset(ps, fd, op, events);

#else /* !ERTS_POLL_USE_CONCURRENT_UPDATE */
    if (fd >= ps->fds_status_len)
	grow_fds_status(ps, fd);

    ASSERT(fd < ps->fds_status_len);

    if (op == ERTS_POLL_OP_DEL) {
        ps->fds_status[fd].flags |= ERTS_POLL_FD_FLG_RST;
        ps->fds_status[fd].events = 0;
        *do_wake = 1;
    } else if (op == ERTS_POLL_OP_ADD) {
        ASSERT(ps->fds_status[fd].events == 0);
        ps->fds_status[fd].events = events;
        *do_wake = 1;
    } else {
        ASSERT(op == ERTS_POLL_OP_MOD);
        ps->fds_status[fd].events = events;
        *do_wake = 1;
    }
    new_events = ps->fds_status[fd].events;

    enqueue_update_request(ps, fd);

#endif /* !ERTS_POLL_USE_CONCURRENT_UPDATE */

 done:
    DEBUG_PRINT_FD("%s = %s(%p, %d, %s, %s) do_wake=%d",
                   ps, fd, ev2str(new_events), __FUNCTION__, ps,
                   fd, op2str(op), ev2str(events), *do_wake);
    return new_events;
}

ErtsPollEvents
ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet *ps,
				    ErtsSysFdType fd,
                                    ErtsPollOp op,
				    ErtsPollEvents events,
				    int* do_wake) /* In: Wake up polling thread */
				                  /* Out: Poller is woken */
{
    ErtsPollEvents res;

    ERTS_POLLSET_LOCK(ps);

    res = poll_control(ps, fd, op, events, do_wake);

    ERTS_POLLSET_UNLOCK(ps);

#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    if (*do_wake) {
	wake_poller(ps, 0);
    }
#endif

    return res;
}

/*
 * --- Wait on poll set ------------------------------------------------------
 */

#if ERTS_POLL_USE_KERNEL_POLL

static ERTS_INLINE int
ERTS_POLL_EXPORT(save_result)(ErtsPollSet *ps, ErtsPollResFd pr[], int max_res, int chk_fds_res, int ebadf)
{
#if !ERTS_POLL_USE_CONCURRENT_UPDATE || ERTS_POLL_DEBUG_PRINT
    int n = chk_fds_res < max_res ? chk_fds_res : max_res, i;
    int res = n;
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    int wake_fd = ps->wake_fds[0];
#endif

    for (i = 0; i < n; i++) {
        int fd = ERTS_POLL_RES_GET_FD(&pr[i]);
#ifdef DEBUG_PRINT_MODE
        ErtsPollEvents evts = ERTS_POLL_RES_GET_EVTS(pr+i);
#endif

        DEBUG_PRINT_FD("trig %s (%s)", ps, fd,
                       ev2str(evts),
#if ERTS_POLL_USE_KQUEUE
                       "kqueue"
#elif ERTS_POLL_USE_EPOLL
                       "epoll"
#else
                       "/dev/poll"
#endif
            );

#if !ERTS_POLL_USE_CONCURRENT_UPDATE

        if (fd == wake_fd) {
            cleanup_wakeup_pipe(ps);
            ERTS_POLL_RES_SET_EVTS(&pr[i], ERTS_POLL_EV_NONE);
        } else {
            /* Reset the events to emulate ONESHOT semantics */
            ps->fds_status[fd].events = 0;
            enqueue_update_request(ps, fd);
        }
#endif
    }

    return res;
#else
    ASSERT(chk_fds_res <= max_res);
    return chk_fds_res;
#endif
}

#else /* !ERTS_POLL_USE_KERNEL_POLL */

static ERTS_INLINE int
ERTS_POLL_EXPORT(save_result)(ErtsPollSet *ps, ErtsPollResFd pr[], int max_res, int chk_fds_res, int ebadf)
{
#if ERTS_POLL_USE_POLL	/* --- poll -------------------------------- */
	int res = 0;
	int wake_fd = ps->wake_fds[0];
	int i, first_ix, end_ix;

	/*
	 * In order to be somewhat fair, we continue on the poll_fds
	 * index where we stopped last time.
	 */
	first_ix = i = ((ps->next_poll_fds_ix < ps->no_poll_fds)
			? ps->next_poll_fds_ix
			: 0);
	end_ix = ps->no_poll_fds;

	while (1) {
	    while (i < end_ix && res < max_res) {
		if (ps->poll_fds[i].revents != (short) 0) {
		    int fd = ps->poll_fds[i].fd;
		    ErtsPollEvents revents;
		    if (fd == wake_fd) {
			cleanup_wakeup_pipe(ps);
			i++;
			continue;
		    }
		    revents = pollev2ev(ps->poll_fds[i].revents);
		    ERTS_POLL_RES_SET_FD(&pr[res], fd);
		    ERTS_POLL_RES_SET_EVTS(&pr[res], revents);

                    /* If an fd returns as error, we may want to check the
                       update_requests queue to see if it has been reset
                       before delivering the result?!?! This should allow
                       the user to do driver_dselect + close without waiting
                       for stop_select... */

                    DEBUG_PRINT_FD("trig %s (poll)", ps, ERTS_POLL_RES_GET_FD(&pr[res]),
                                   ev2str(ERTS_POLL_RES_GET_EVTS(&pr[res])));

		    res++;

                    /* Clear the events for this fd in order to mimic
                       how epoll ONESHOT works */
                    ps->fds_status[fd].events = 0;
                    enqueue_update_request(ps, fd);
		}
		i++;
	    }
	    if (res == max_res || i == first_ix)
		break;
	    ASSERT(i == ps->no_poll_fds);
	    i = 0;
	    end_ix = first_ix;
	}

	ps->next_poll_fds_ix = i;
	return res;

#elif ERTS_POLL_USE_SELECT	/* --- select ------------------------------ */
	int res = 0;
	int wake_fd = ps->wake_fds[0];
	int fd, first_fd, end_fd;

	/*
	 * In order to be fair, we continue on the fd where we stopped
	 * last time.
	 */
	first_fd = fd = ps->next_sel_fd <= ps->max_fd ? ps->next_sel_fd : 0;
	end_fd = ps->max_fd + 1;

	if (!ebadf) {
	    while (1) {
		while (fd < end_fd && res < max_res) {
                    ErtsPollEvents events = 0;
		    if (ERTS_FD_ISSET(fd, &ps->res_input_fds)) {
			if (fd == wake_fd) {
			    cleanup_wakeup_pipe(ps);
			    fd++;
			    continue;
			}
			events |= ERTS_POLL_EV_IN;
		    }
		    if (ERTS_FD_ISSET(fd, &ps->res_output_fds))
			events |= ERTS_POLL_EV_OUT;
		    if (events) {
                        ERTS_POLL_RES_SET_FD(&pr[res], fd);
                        ERTS_POLL_RES_SET_EVTS(&pr[res], events);
			res++;
                        ps->fds_status[fd].events = 0;
                        enqueue_update_request(ps, fd);
		    }
		    fd++;
		}
		if (res == max_res || fd == first_fd)
		    break;
		ASSERT(fd == ps->max_fd + 1);
		fd = 0;
		end_fd = first_fd;
	    }
	}
	else {
	    /*
	     * Bad file descriptors in poll set.
	     *
	     * This only happens when running poorly written
	     * drivers. This code could be optimized, but we
	     * don't bother since it should never happen...
	     */
	    while (1) {
		while (fd < end_fd && res < max_res) {
		    if (ps->fds_status[fd].events) {
			int sres;
			ERTS_fd_set *iset = NULL;
			ERTS_fd_set *oset = NULL;
			if (ps->fds_status[fd].events & ERTS_POLL_EV_IN) {
			    iset = &ps->res_input_fds;
			    ERTS_FD_ZERO(iset);
			    ERTS_FD_SET(fd, iset);
			}
			if (ps->fds_status[fd].events & ERTS_POLL_EV_OUT) {
			    oset = &ps->res_output_fds;
			    ERTS_FD_ZERO(oset);
			    ERTS_FD_SET(fd, oset);
			}
			do {
			    /* Initiate 'tv' each time;
			       select() may modify it */
			    SysTimeval tv = {0, 0};
			    sres = ERTS_SELECT(ps->max_fd+1, iset, oset, NULL, &tv);
			} while (sres < 0 && errno == EINTR);
			if (sres < 0) {
			    if (fd == wake_fd) {
				cleanup_wakeup_pipe(ps);
				fd++;
				continue;
			    }
                            ERTS_POLL_RES_SET_FD(&pr[res], fd);
                            ERTS_POLL_RES_SET_EVTS(&pr[res], ERTS_POLL_EV_NVAL);
			    res++;
			}
			else if (sres > 0) {
                            ErtsPollEvents events = 0;
                            ERTS_POLL_RES_SET_FD(&pr[res], fd);
			    if (iset && ERTS_FD_ISSET(fd, iset)) {
				if (fd == wake_fd) {
				    cleanup_wakeup_pipe(ps);
				    fd++;
				    continue;
				}
				events |= ERTS_POLL_EV_IN;
			    }
			    if (oset && ERTS_FD_ISSET(fd, oset)) {
				events |= ERTS_POLL_EV_OUT;
			    }
			    ASSERT(events);
                            ERTS_POLL_RES_SET_EVTS(&pr[res], events);
			    res++;
			}
		    }
		    fd++;
		}
		if (res == max_res || fd == first_fd)
		    break;
		ASSERT(fd == ps->max_fd + 1);
		fd = 0;
		end_fd = first_fd;
	    }
	}
	ps->next_sel_fd = fd;
	return res;
#endif /* ERTS_POLL_USE_SELECT */
}

#endif /* !ERTS_POLL_USE_KERNEL_POLL */

static ERTS_INLINE int
check_fd_events(ErtsPollSet *ps, ErtsPollResFd pr[], int do_wait, int max_res)
{
    int res;
    int timeout = do_wait ? -1 : 0;
    DEBUG_PRINT_WAIT("Entering check_fd_events(), do_wait=%d", ps, do_wait);
    {
#if ERTS_POLL_USE_EPOLL                /* --- epoll ------------------------------- */
        res = epoll_wait(ps->kp_fd, pr, max_res, timeout);

#elif ERTS_POLL_USE_KQUEUE     /* --- kqueue ------------------------------ */
        struct timespec ts = {0, 0};
        struct timespec *tsp = timeout ? NULL : &ts;
        res = kevent(ps->kp_fd, NULL, 0, pr, max_res, tsp);
#elif ERTS_POLL_USE_DEVPOLL	/* --- devpoll ----------------------------- */
        /*
         * The ioctl() will fail with EINVAL on Solaris 10 if dp_nfds
         * is set too high. dp_nfds should not be set greater than
         * the maximum number of file descriptors in the poll set.
         */
        struct dvpoll poll_res;
        int nfds = (int) erts_atomic_read_nob(&ps->no_of_user_fds) + 1 /* wakeup pipe */;
        poll_res.dp_nfds = nfds < max_res ? nfds : max_res;
        poll_res.dp_fds = pr;
        poll_res.dp_timeout = timeout;
        res = ioctl(ps->kp_fd, DP_POLL, &poll_res);

#elif ERTS_POLL_USE_POLL        /* --- poll --------------------------------- */

        res = poll(ps->poll_fds, ps->no_poll_fds, timeout);

#elif ERTS_POLL_USE_SELECT	/* --- select ------------------------------ */
        SysTimeval tv = {0, 0};
        SysTimeval *tvp = timeout ? NULL : &tv;

        ERTS_FD_COPY(&ps->input_fds, &ps->res_input_fds);
        ERTS_FD_COPY(&ps->output_fds, &ps->res_output_fds);

        res = ERTS_SELECT(ps->max_fd + 1,
                          &ps->res_input_fds,
                          &ps->res_output_fds,
                          NULL,
                          tvp);
#endif				/* ----------------------------------------- */
    }
    DEBUG_PRINT_WAIT("Leaving check_fd_events(), res=%d", ps, res);
    return res;
}

int
ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet *ps,
				 ErtsPollResFd pr[],
				 int *len)
{
    int res, no_fds, used_fds = 0;
    int ebadf = 0;
    int do_wait;
    int ps_locked = 0;
    ERTS_MSACC_DECLARE_CACHE();

    no_fds = *len;
    *len = 0;
    ASSERT(no_fds > 0);

#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    if (ERTS_POLLSET_HAVE_UPDATE_REQUESTS(ps)) {
	ERTS_POLLSET_LOCK(ps);
	used_fds = handle_update_requests(ps, pr, no_fds);
	ERTS_POLLSET_UNLOCK(ps);

        if (used_fds == no_fds) {
            *len = used_fds;
            return 0;
        }
    }
#endif

    do_wait = !is_woken(ps) && used_fds == 0;

    DEBUG_PRINT_WAIT("Entering %s(), do_wait=%d", ps, __FUNCTION__, do_wait);

    if (do_wait) {
        erts_thr_progress_prepare_wait(NULL);
        ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_SLEEP);
    }

    while (1) {
	res = check_fd_events(ps, pr + used_fds, do_wait, no_fds - used_fds);

#if !ERTS_POLL_USE_CONCURRENT_UPDATE
        if (res < 0
            && errno == EBADF
            && ERTS_POLLSET_HAVE_UPDATE_REQUESTS(ps)) {
            /*
             * This may have happened because another thread deselected
             * a fd in our poll set and then closed it, i.e. the driver
             * behaved correctly. We wan't to avoid looking for a bad
             * fd, that may even not exist anymore. Therefore, handle
             * update requests and try again. This behaviour should only
             * happen when using SELECT as the polling mechanism.
             */
            ERTS_POLLSET_LOCK(ps);
            used_fds += handle_update_requests(ps, pr + used_fds, no_fds - used_fds);
            if (used_fds == no_fds) {
                *len = used_fds;
                ERTS_POLLSET_UNLOCK(ps);
                return 0;
            }
            res = check_fd_events(ps, pr + used_fds, 0, no_fds - used_fds);
            /* Keep the lock over the non-blocking poll in order to not
               get any nasty races happening. */
            ERTS_POLLSET_UNLOCK(ps);
            if (res == 0) {
                errno = EAGAIN;
                res = -1;
            }
        }
#endif

	if (res != 0)
	    break;
        if (!do_wait)
            break;
    }

    if (do_wait) {
        erts_thr_progress_finalize_wait(NULL);
        ERTS_MSACC_UPDATE_CACHE();
        ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_CHECK_IO);
    }

    woke_up(ps);

    if (res < 0) {
#if ERTS_POLL_USE_SELECT
	if (errno == EBADF) {
	    ebadf = 1;
	    goto save_results;
	}
#endif
	res = errno;
    }
    else {
#if ERTS_POLL_USE_SELECT
    save_results:
#endif

	ps_locked = 1;
	ERTS_POLLSET_LOCK(ps);

	used_fds += ERTS_POLL_EXPORT(save_result)(ps, pr + used_fds, no_fds - used_fds, res, ebadf);

#ifdef HARD_DEBUG
	check_poll_result(pr, used_fds);
#endif

	res = (used_fds == 0 ? (is_interrupted_reset(ps) ? EINTR : EAGAIN) : 0);
	*len = used_fds;
    }

    if (ps_locked)
	ERTS_POLLSET_UNLOCK(ps);

    DEBUG_PRINT_WAIT("Leaving %s = %s(len = %d)", ps,
                     res == 0 ? "0" : erl_errno_id(res), __FUNCTION__, *len);

    return res;
}

/*
 * --- Interrupt a thread doing erts_poll_wait() -----------------------------
 */

void
ERTS_POLL_EXPORT(erts_poll_interrupt)(ErtsPollSet *ps, int set)
{
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    if (!set)
	reset_wakeup_state(ps);
    else
	wake_poller(ps, 1);
#endif
}

int
ERTS_POLL_EXPORT(erts_poll_max_fds)(void)
{
    return max_fds;
}
/*
 * --- Initialization --------------------------------------------------------
 */

void
ERTS_POLL_EXPORT(erts_poll_init)(int *concurrent_updates)
{

    errno = 0;

    if (concurrent_updates) {
#if ERTS_POLL_USE_CONCURRENT_UPDATE
        *concurrent_updates = 1;
#else
        *concurrent_updates = 0;
#endif
    }

#if !defined(NO_SYSCONF)
    max_fds = sysconf(_SC_OPEN_MAX);
#elif ERTS_POLL_USE_SELECT
    max_fds = NOFILE;
#else
    max_fds = OPEN_MAX;
#endif

#if ERTS_POLL_USE_SELECT && defined(FD_SETSIZE) && \
	!defined(_DARWIN_UNLIMITED_SELECT)
    if (max_fds > FD_SETSIZE)
	max_fds = FD_SETSIZE;
#endif

    if (max_fds < 0)
	fatal_error("erts_poll_init(): Failed to get max number of files: %s\n",
		    erl_errno_id(errno));

    print_misc_debug_info();
}

ErtsPollSet *
ERTS_POLL_EXPORT(erts_poll_create_pollset)(int id)
{
#if ERTS_POLL_USE_KERNEL_POLL
    int kp_fd;
#endif
    ErtsPollSet *ps = erts_alloc(ERTS_ALC_T_POLLSET,
				sizeof(struct ERTS_POLL_EXPORT(erts_pollset)));
    ps->id = id;
    ps->internal_fd_limit = 0;
    erts_atomic_init_nob(&ps->no_of_user_fds, 0);
#if ERTS_POLL_USE_KERNEL_POLL
    ps->kp_fd = -1;
#if ERTS_POLL_USE_EPOLL
    kp_fd = epoll_create(256);
#elif ERTS_POLL_USE_DEVPOLL
    kp_fd = open("/dev/poll", O_RDWR);
#elif ERTS_POLL_USE_KQUEUE
    kp_fd = kqueue();
#endif
    if (kp_fd < 0)
	fatal_error("erts_poll_create_pollset(): Failed to "
#if ERTS_POLL_USE_EPOLL
		    "create epoll set"
#elif ERTS_POLL_USE_DEVPOLL
		    "to open /dev/poll"
#elif ERTS_POLL_USE_KQUEUE
		    "create kqueue"
#endif
		    ": %s (%d)\n",
		    erl_errno_id(errno), errno);
#endif /* ERTS_POLL_USE_KERNEL_POLL */
#if ERTS_POLL_USE_POLL
    ps->next_poll_fds_ix = 0;
    ps->no_poll_fds = 0;
    ps->poll_fds_len = 0;
    ps->poll_fds = NULL;
#elif ERTS_POLL_USE_SELECT
    ps->next_sel_fd = 0;
    ps->max_fd = -1;
#ifdef _DARWIN_UNLIMITED_SELECT
    ps->input_fds.sz = 0;
    ps->input_fds.ptr = NULL;
    ps->res_input_fds.sz = 0;
    ps->res_input_fds.ptr = NULL;
    ps->output_fds.sz = 0;
    ps->output_fds.ptr = NULL;
    ps->res_output_fds.sz = 0;
    ps->res_output_fds.ptr = NULL;
#else
    ERTS_FD_ZERO(&ps->input_fds);
    ERTS_FD_ZERO(&ps->res_input_fds);
    ERTS_FD_ZERO(&ps->output_fds);
    ERTS_FD_ZERO(&ps->res_output_fds);
#endif
#endif
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    ps->fds_status = NULL;
    ps->fds_status_len = 0;
    ps->update_requests.next = NULL;
    ps->update_requests.len = 0;
    ps->curr_upd_req_block = &ps->update_requests;
    erts_atomic32_init_nob(&ps->have_update_requests, 0);
    erts_mtx_init(&ps->mtx, "pollset", NIL, ERTS_LOCK_FLAGS_CATEGORY_IO);
#endif
#if ERTS_POLL_USE_KERNEL_POLL
    if (ps->internal_fd_limit <= kp_fd)
	ps->internal_fd_limit = kp_fd + 1;
    ps->kp_fd = kp_fd;
#endif
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    erts_atomic32_init_nob(&ps->wakeup_state, (erts_aint32_t) 0);
    create_wakeup_pipe(ps);
    handle_update_requests(ps, NULL, 0);
    cleanup_wakeup_pipe(ps);
#endif
    erts_atomic_set_nob(&ps->no_of_user_fds, 0); /* Don't count wakeup pipe and fallback fd */

    return ps;
}

/*
 * --- Info ------------------------------------------------------------------
 */

void
ERTS_POLL_EXPORT(erts_poll_info)(ErtsPollSet *ps, ErtsPollInfo *pip)
{
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    int pending_updates;
#endif
    Uint size = 0;

    ERTS_POLLSET_LOCK(ps);

    size += sizeof(struct ERTS_POLL_EXPORT(erts_pollset));
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    size += ps->fds_status_len*sizeof(ErtsFdStatus);
#endif

#if ERTS_POLL_USE_POLL
    size += ps->poll_fds_len*sizeof(struct pollfd);
#elif ERTS_POLL_USE_SELECT
#ifdef _DARWIN_UNLIMITED_SELECT
    size += ps->input_fds.sz + ps->res_input_fds.sz
	 + ps->output_fds.sz + ps->res_output_fds.sz;
#endif
#endif

#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    {
	ErtsPollSetUpdateRequestsBlock *urqbp = ps->update_requests.next;
	pending_updates = ps->update_requests.len;
	while (urqbp) {
	    size += sizeof(ErtsPollSetUpdateRequestsBlock);
	    pending_updates += urqbp->len;
	    urqbp = urqbp->next;
	}
    }
#endif

    pip->primary =
#if ERTS_POLL_USE_KQUEUE
	"kqueue"
#elif ERTS_POLL_USE_EPOLL
	"epoll"
#elif ERTS_POLL_USE_DEVPOLL
	"/dev/poll"
#elif ERTS_POLL_USE_POLL
	"poll"
#elif ERTS_POLL_USE_SELECT
	"select"
#endif
	;

    pip->kernel_poll =
#if !ERTS_POLL_USE_KERNEL_POLL
	NULL
#elif ERTS_POLL_USE_KQUEUE
	"kqueue"
#elif ERTS_POLL_USE_EPOLL
	"epoll"
#elif ERTS_POLL_USE_DEVPOLL
	"/dev/poll"
#endif
	;

    pip->memory_size = size;

    pip->poll_set_size = (int) erts_atomic_read_nob(&ps->no_of_user_fds);
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    pip->poll_set_size++; /* Wakeup pipe */
#endif

    pip->lazy_updates =
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
	1
#else
	0
#endif
	;

    pip->pending_updates = 
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
	pending_updates
#else
	0
#endif
	;

    pip->concurrent_updates =
#if ERTS_POLL_USE_CONCURRENT_UPDATE
	1
#else
	0
#endif
	;

    pip->is_fallback =
#if ERTS_POLL_IS_FALLBACK
        1
#else
        0
#endif
        ;

    pip->max_fds = max_fds;

    ERTS_POLLSET_UNLOCK(ps);

}

/*
 * Fatal error...
 */

#ifndef ERTS_GOT_SIGUSR1
#  define ERTS_GOT_SIGUSR1 0
#endif

static void
fatal_error(char *format, ...)
{
    va_list ap;

    if (ERTS_SOMEONE_IS_CRASH_DUMPING || ERTS_GOT_SIGUSR1) {
	/*
	 * Crash dump writing and reception of sigusr1 (which will
	 * result in a crash dump) closes all file descriptors. This
	 * typically results in a fatal error for erts_poll() (wakeup
	 * pipes and kernel poll fds are closed).
	 *
	 * We ignore the error and let the crash dump writing continue...
	 */
	return;
    }
    va_start(ap, format);
    erts_vfprintf(stderr, format, ap);
    va_end(ap);
    abort();
}

/*
 * --- Debug -----------------------------------------------------------------
 */

#if ERTS_POLL_USE_EPOLL
uint32_t epoll_events(int kp_fd, int fd)
{
    /* For epoll we read the information about what is selected upon from the proc fs.*/
    char fname[30];
    FILE *f;
    unsigned int pos, flags, mnt_id;
    int line = 0;
    sprintf(fname,"/proc/%d/fdinfo/%d",getpid(), kp_fd);
    f = fopen(fname,"r");
    if (!f) {
        fprintf(stderr,"failed to open file %s, errno = %d\n", fname, errno);
        ASSERT(0);
        return 0;
    }
    if (fscanf(f,"pos:\t%x\nflags:\t%x", &pos, &flags) != 2) {
        fprintf(stderr,"failed to parse file %s, errno = %d\n", fname, errno);
        ASSERT(0);
        return 0;
    }
    if (fscanf(f,"\nmnt_id:\t%x\n", &mnt_id));
    line += 3;
    while (!feof(f)) {
        /* tfd:       10 events: 40000019 data:       180000000a */
        int ev_fd;
        uint32_t events;
        uint64_t data;
        if (fscanf(f,"tfd:%d events:%x data:%lx\n", &ev_fd, &events, &data) != 3) {
            fprintf(stderr,"failed to parse file %s on line %d, errno = %d\n", fname,
                    line,
                    errno);
            return 0;
        }
        if (fd == ev_fd) {
            fclose(f);
            return events;
        }
    }
    fclose(f);
    return 0;
}
#endif

void
ERTS_POLL_EXPORT(erts_poll_get_selected_events)(ErtsPollSet *ps,
						ErtsPollEvents ev[],
						int len)
{
    int fd;
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    ERTS_POLLSET_LOCK(ps);
    for (fd = 0; fd < len; fd++) {
	if (fd >= ps->fds_status_len)
	    ev[fd] = 0;
	else {
	    ev[fd] = ps->fds_status[fd].events;
            if (
                fd == ps->wake_fds[0] || fd == ps->wake_fds[1] ||
#if ERTS_POLL_USE_KERNEL_POLL
                fd == ps->kp_fd ||
#endif
                0)
                ev[fd] |= ERTS_POLL_EV_NVAL;
	}
    }
    ERTS_POLLSET_UNLOCK(ps);
#elif ERTS_POLL_USE_EPOLL

    /* For epoll we read the information about what is selected upon from the proc fs.*/
    char fname[30];
    FILE *f;
    unsigned int pos, flags, mnt_id;
    int line = 0;
    sprintf(fname,"/proc/%d/fdinfo/%d",getpid(), ps->kp_fd);
    for (fd = 0; fd < len; fd++)
        ev[fd] = ERTS_POLL_EV_NONE;
    f = fopen(fname,"r");
    if (!f) {
        fprintf(stderr,"failed to open file %s, errno = %d\n", fname, errno);
        return;
    }
    if (fscanf(f,"pos:\t%x\nflags:\t%x", &pos, &flags) != 2) {
        fprintf(stderr,"failed to parse file %s, errno = %d\n", fname, errno);
        ASSERT(0);
        return;
    }
    if (fscanf(f,"\nmnt_id:\t%x\n", &mnt_id));
    line += 3;
    while (!feof(f)) {
        /* tfd:       10 events: 40000019 data:       180000000a */
        int fd;
        uint32_t events;
        uint64_t data;
        if (fscanf(f,"tfd:%d events:%x data:%lx\n", &fd, &events, &data) != 3) {
            fprintf(stderr,"failed to parse file %s on line %d, errno = %d\n",
                    fname, line, errno);
            ASSERT(0);
            return;
        }
        data &= 0xFFFFFFFF;
        ASSERT(fd == data);
        /* Events are the events that are being monitored, which of course include
           error and hup events, but we are only interested in IN/OUT events */
        ev[fd] = (ERTS_POLL_EV_IN|ERTS_POLL_EV_OUT) & ERTS_POLL_EV_N2E(events);
        line++;
    }
#else
    for (fd = 0; fd < len; fd++)
        ev[fd] = ERTS_POLL_EV_NONE;
#endif
}

#ifdef HARD_DEBUG

static void
check_poll_result(ErtsPollResFd pr[], int len)
{
    int i, j;

    for (i = 0; i < len; i++) {
	ASSERT(pr[i].fd >= 0);
	ASSERT(pr[i].fd < max_fds);
	for (j = 0; j < len; j++) {
	    ASSERT(i == j || pr[i].fd != pr[j].fd);
	}
    }
}


#if ERTS_POLL_USE_DEVPOLL && defined(DEBUG)

static void
check_poll_status(ErtsPollSet *ps)
{
    int i;
    for (i = 0; i < ps->fds_status_len; i++) {
	int ires;
	struct pollfd dp_fd;
	short events = ERTS_POLL_EV_E2N(ps->fds_status[i].events);
	
	dp_fd.fd = i;
	dp_fd.events = (short) 0;
	dp_fd.revents = (short) 0;

	ires = ioctl(ps->kp_fd, DP_ISPOLLED, &dp_fd);

	if (ires == 0) {
	    ASSERT(!events);
	}
	else if (ires == 1) {
	    ASSERT(events);
	    ASSERT(events == dp_fd.revents);
	}
	else {
	    ASSERT(0);
	}
	ASSERT(dp_fd.fd == i);
	ASSERT(ps->fds_status[i].events == ps->fds_status[i].used_events);
    }
}

#endif /* ERTS_POLL_USE_DEVPOLL */
#endif /* HARD_DEBUG */

static void
print_misc_debug_info(void)
{
#if ERTS_POLL_DEBUG_PRINT
    erts_printf("erts_poll using: %s lazy_updates:%s\n",
#if ERTS_POLL_USE_KQUEUE
		"kqueue"
#elif ERTS_POLL_USE_EPOLL
		"epoll"
#elif ERTS_POLL_USE_DEVPOLL
		"/dev/poll"
#elif ERTS_POLL_USE_POLL
		"poll"
#elif ERTS_POLL_USE_SELECT
		"select"
#endif
		,
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
		"true"
#else
		"false"
#endif
		);

    erts_printf("ERTS_POLL_EV_IN=0x%x\n"
		"ERTS_POLL_EV_OUT=0x%x\n"
		"ERTS_POLL_EV_NVAL=0x%x\n"
		"ERTS_POLL_EV_ERR=0x%x\n",
		ERTS_POLL_EV_IN,
		ERTS_POLL_EV_OUT,
		ERTS_POLL_EV_NVAL,
		ERTS_POLL_EV_ERR);

#ifdef FD_SETSIZE
    erts_printf("FD_SETSIZE=%d\n", FD_SETSIZE);
#endif
#endif
}

#ifdef ERTS_ENABLE_LOCK_COUNT
void ERTS_POLL_EXPORT(erts_lcnt_enable_pollset_lock_count)(ErtsPollSet *pollset, int enable)
{
#if !ERTS_POLL_USE_CONCURRENT_UPDATE
    if(enable) {
        erts_lcnt_install_new_lock_info(&pollset->mtx.lcnt, "pollset_rm", NIL,
            ERTS_LOCK_TYPE_MUTEX | ERTS_LOCK_FLAGS_CATEGORY_IO);
    } else {
        erts_lcnt_uninstall(&pollset->mtx.lcnt);
    }
#endif
    return;
}
#endif