summaryrefslogtreecommitdiff
path: root/rts/Schedule.c
blob: 9b1e98ea179438c47e31e65fa7bef69623e7c9f6 (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
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
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
/* ---------------------------------------------------------------------------
 *
 * (c) The GHC Team, 1998-2006
 *
 * The scheduler and thread-related functionality
 *
 * --------------------------------------------------------------------------*/

#include "rts/PosixSource.h"
#define KEEP_LOCKCLOSURE
#include "Rts.h"

#include "sm/Storage.h"
#include "RtsUtils.h"
#include "StgRun.h"
#include "Schedule.h"
#include "Interpreter.h"
#include "Printer.h"
#include "RtsSignals.h"
#include "sm/Sanity.h"
#include "Stats.h"
#include "STM.h"
#include "Prelude.h"
#include "ThreadLabels.h"
#include "Updates.h"
#include "Proftimer.h"
#include "ProfHeap.h"
#include "Weak.h"
#include "sm/GC.h" // waitForGcThreads, releaseGCThreads, N
#include "sm/GCThread.h"
#include "Sparks.h"
#include "Capability.h"
#include "Task.h"
#include "AwaitEvent.h"
#include "IOManager.h"
#if defined(mingw32_HOST_OS)
#include "win32/MIOManager.h"
#include "win32/AsyncWinIO.h"
#endif
#include "Trace.h"
#include "RaiseAsync.h"
#include "Threads.h"
#include "Timer.h"
#include "ThreadPaused.h"
#include "Messages.h"
#include "StablePtr.h"
#include "StableName.h"
#include "TopHandler.h"
#include "sm/NonMoving.h"
#include "sm/NonMovingMark.h"

#if defined(HAVE_SYS_TYPES_H)
#include <sys/types.h>
#endif
#if defined(HAVE_UNISTD_H)
#include <unistd.h>
#endif

#include <string.h>
#include <stdlib.h>
#include <stdarg.h>

#if defined(HAVE_ERRNO_H)
#include <errno.h>
#endif

#if defined(TRACING)
#include "eventlog/EventLog.h"
#endif
/* -----------------------------------------------------------------------------
 * Global variables
 * -------------------------------------------------------------------------- */

#if !defined(THREADED_RTS)
// Blocked/sleeping threads
StgTSO *blocked_queue_hd = NULL;
StgTSO *blocked_queue_tl = NULL;
StgTSO *sleeping_queue = NULL;    // perhaps replace with a hash table?
#endif

// Bytes allocated since the last time a HeapOverflow exception was thrown by
// the RTS
uint64_t allocated_bytes_at_heapoverflow = 0;

/* Set to true when the latest garbage collection failed to reclaim enough
 * space, and the runtime should proceed to shut itself down in an orderly
 * fashion (emitting profiling info etc.), OR throw an exception to the main
 * thread, if it is still alive.
 */
bool heap_overflow = false;

/* flag that tracks whether we have done any execution in this time slice.
 * LOCK: currently none, perhaps we should lock (but needs to be
 * updated in the fast path of the scheduler).
 *
 * NB. must be StgWord, we do xchg() on it.
 */
volatile StgWord recent_activity = ACTIVITY_YES;

/* if this flag is set as well, give up execution
 * LOCK: none (changes monotonically)
 */
volatile StgWord sched_state = SCHED_RUNNING;

/*
 * This mutex protects most of the global scheduler data in
 * the THREADED_RTS runtime.
 */
#if defined(THREADED_RTS)
Mutex sched_mutex;
#endif

#if !defined(mingw32_HOST_OS)
#define FORKPROCESS_PRIMOP_SUPPORTED
#endif

/*
 * sync_finished_cond allows threads which do not own any capability (e.g. the
 * concurrent mark thread) to participate in the sync protocol. In particular,
 * if such a thread requests a sync while sync is already in progress it will
 * block on sync_finished_cond, which will be signalled when the sync is
 * finished (by releaseAllCapabilities).
 */
#if defined(THREADED_RTS)
static Condition sync_finished_cond;
static Mutex sync_finished_mutex;
#endif


/* -----------------------------------------------------------------------------
 * static function prototypes
 * -------------------------------------------------------------------------- */

static Capability *schedule (Capability *initialCapability, Task *task);

//
// These functions all encapsulate parts of the scheduler loop, and are
// abstracted only to make the structure and control flow of the
// scheduler clearer.
//
static void scheduleFindWork (Capability **pcap);
#if defined(THREADED_RTS)
static void scheduleYield (Capability **pcap, Task *task);
#endif
#if defined(THREADED_RTS)
static bool requestSync (Capability **pcap, Task *task,
                         PendingSync *sync_type, SyncType *prev_sync_type);
static void acquireAllCapabilities(Capability *cap, Task *task);
static void startWorkerTasks (uint32_t from USED_IF_THREADS,
                              uint32_t to USED_IF_THREADS);
#endif
static void scheduleStartSignalHandlers (Capability *cap);
static void scheduleCheckBlockedThreads (Capability *cap);
static void scheduleProcessInbox(Capability **cap);
static void scheduleDetectDeadlock (Capability **pcap, Task *task);
static void schedulePushWork(Capability *cap, Task *task);
#if defined(THREADED_RTS)
static void scheduleActivateSpark(Capability *cap);
#endif
static void schedulePostRunThread(Capability *cap, StgTSO *t);
static bool scheduleHandleHeapOverflow( Capability *cap, StgTSO *t );
static bool scheduleHandleYield( Capability *cap, StgTSO *t,
                                 uint32_t prev_what_next );
static void scheduleHandleThreadBlocked( StgTSO *t );
static bool scheduleHandleThreadFinished( Capability *cap, Task *task,
                                          StgTSO *t );
static bool scheduleNeedHeapProfile(bool ready_to_gc);
static void scheduleDoGC( Capability **pcap, Task *task,
                          bool force_major, bool is_overflow_gc, bool deadlock_detect );

static void deleteThread (StgTSO *tso);
static void deleteAllThreads (void);

#if defined(FORKPROCESS_PRIMOP_SUPPORTED)
static void deleteThread_(StgTSO *tso);
#endif

/* ---------------------------------------------------------------------------
   Main scheduling loop.

   We use round-robin scheduling, each thread returning to the
   scheduler loop when one of these conditions is detected:

      * out of heap space
      * timer expires (thread yields)
      * thread blocks
      * thread ends
      * stack overflow

   ------------------------------------------------------------------------ */

static Capability *
schedule (Capability *initialCapability, Task *task)
{
  StgTSO *t;
  Capability *cap;
  StgThreadReturnCode ret;
  uint32_t prev_what_next;
  bool ready_to_gc;

  cap = initialCapability;
  t = NULL;

  // Pre-condition: this task owns initialCapability.
  // The sched_mutex is *NOT* held
  // NB. on return, we still hold a capability.
  ASSERT_FULL_CAPABILITY_INVARIANTS(cap, task);

  debugTrace (DEBUG_sched, "cap %d: schedule()", initialCapability->no);

  // -----------------------------------------------------------
  // Scheduler loop starts here:

  while (1) {
    ASSERT_FULL_CAPABILITY_INVARIANTS(cap, task);

    // Check whether we have re-entered the RTS from Haskell without
    // going via suspendThread()/resumeThread (i.e. a 'safe' foreign
    // call).
    if (cap->in_haskell) {
          errorBelch("schedule: re-entered unsafely.\n"
                     "   Perhaps a 'foreign import unsafe' should be 'safe'?");
          stg_exit(EXIT_FAILURE);
    }

    // Note [shutdown]: The interruption / shutdown sequence.
    // ~~~~~~~~~~~~~~~
    //
    // In order to cleanly shut down the runtime, we want to:
    //   * make sure that all main threads return to their callers
    //     with the state 'Interrupted'.
    //   * clean up all OS threads associated with the runtime
    //   * free all memory etc.
    //
    // So the sequence goes like this:
    //
    //   * The shutdown sequence is initiated by calling hs_exit(),
    //     interruptStgRts(), or running out of memory in the GC.
    //
    //   * Set sched_state = SCHED_INTERRUPTING
    //
    //   * The scheduler notices sched_state = SCHED_INTERRUPTING and calls
    //     scheduleDoGC(), which halts the whole runtime by acquiring all the
    //     capabilities, does a GC and then calls deleteAllThreads() to kill all
    //     the remaining threads.  The zombies are left on the run queue for
    //     cleaning up.  We can't kill threads involved in foreign calls.
    //
    //   * scheduleDoGC() sets sched_state = SCHED_SHUTTING_DOWN
    //
    //   * After this point, there can be NO MORE HASKELL EXECUTION.  This is
    //     enforced by the scheduler, which won't run any Haskell code when
    //     sched_state >= SCHED_INTERRUPTING, and we already sync'd with the
    //     other capabilities by doing the GC earlier.
    //
    //   * all workers exit when the run queue on their capability
    //     drains.  All main threads will also exit when their TSO
    //     reaches the head of the run queue and they can return.
    //
    //   * eventually all Capabilities will shut down, and the RTS can
    //     exit.
    //
    //   * We might be left with threads blocked in foreign calls,
    //     we should really attempt to kill these somehow (TODO).

    switch (RELAXED_LOAD(&sched_state)) {
    case SCHED_RUNNING:
        break;
    case SCHED_INTERRUPTING:
        debugTrace(DEBUG_sched, "SCHED_INTERRUPTING");
        /* scheduleDoGC() deletes all the threads */
        scheduleDoGC(&cap,task,true,false,false);

        // after scheduleDoGC(), we must be shutting down.  Either some
        // other Capability did the final GC, or we did it above,
        // either way we can fall through to the SCHED_SHUTTING_DOWN
        // case now.
        ASSERT(RELAXED_LOAD(&sched_state) == SCHED_SHUTTING_DOWN);
        // fall through

    case SCHED_SHUTTING_DOWN:
        debugTrace(DEBUG_sched, "SCHED_SHUTTING_DOWN");
        // If we are a worker, just exit.  If we're a bound thread
        // then we will exit below when we've removed our TSO from
        // the run queue.
        if (!isBoundTask(task) && emptyRunQueue(cap)) {
            return cap;
        }
        break;
    default:
        barf("sched_state: %" FMT_Word, sched_state);
    }

    scheduleFindWork(&cap);

    /* work pushing, currently relevant only for THREADED_RTS:
       (pushes threads, wakes up idle capabilities for stealing) */
    schedulePushWork(cap,task);

    scheduleDetectDeadlock(&cap,task);

    // Normally, the only way we can get here with no threads to
    // run is if a keyboard interrupt received during
    // scheduleCheckBlockedThreads() or scheduleDetectDeadlock().
    // Additionally, it is not fatal for the
    // threaded RTS to reach here with no threads to run.
    //
    // Since IOPorts have no deadlock avoidance guarantees you may also reach
    // this point when blocked on an IO Port.  If this is the case the only
    // thing that could unblock it is an I/O event.
    //
    // win32: might be here due to awaitEvent() being abandoned
    // as a result of a console event having been delivered or as a result of
    // waiting on an async I/O to complete with WinIO.

#if defined(THREADED_RTS)
    scheduleYield(&cap,task);

    if (emptyRunQueue(cap)) continue; // look for work again
#endif

#if !defined(THREADED_RTS)
    if ( emptyRunQueue(cap) ) {
#if defined(mingw32_HOST_OS)
        /* Notify the I/O manager that we have nothing to do.  If there are
           any outstanding I/O requests we'll block here.  If there are not
           then this is a user error and we will abort soon.  */
        awaitEvent (emptyRunQueue(cap));
#else
        ASSERT(sched_state >= SCHED_INTERRUPTING);
#endif
    }
#endif

    //
    // Get a thread to run
    //
    t = popRunQueue(cap);

    // Sanity check the thread we're about to run.  This can be
    // expensive if there is lots of thread switching going on...
    IF_DEBUG(sanity,checkTSO(t));

#if defined(THREADED_RTS)
    // Check whether we can run this thread in the current task.
    // If not, we have to pass our capability to the right task.
    {
        InCall *bound = t->bound;

        if (bound) {
            if (bound->task == task) {
                // yes, the Haskell thread is bound to the current native thread
            } else {
                debugTrace(DEBUG_sched,
                           "thread %lu bound to another OS thread",
                           (unsigned long)t->id);
                // no, bound to a different Haskell thread: pass to that thread
                pushOnRunQueue(cap,t);
                continue;
            }
        } else {
            // The thread we want to run is unbound.
            if (task->incall->tso) {
                debugTrace(DEBUG_sched,
                           "this OS thread cannot run thread %lu",
                           (unsigned long)t->id);
                // no, the current native thread is bound to a different
                // Haskell thread, so pass it to any worker thread
                pushOnRunQueue(cap,t);
                continue;
            }
        }
    }
#endif

    // If we're shutting down, and this thread has not yet been
    // killed, kill it now.  This sometimes happens when a finalizer
    // thread is created by the final GC, or a thread previously
    // in a foreign call returns.
    if (RELAXED_LOAD(&sched_state) >= SCHED_INTERRUPTING &&
        !(t->what_next == ThreadComplete || t->what_next == ThreadKilled)) {
        deleteThread(t);
    }

    // If this capability is disabled, migrate the thread away rather
    // than running it.  NB. but not if the thread is bound: it is
    // really hard for a bound thread to migrate itself.  Believe me,
    // I tried several ways and couldn't find a way to do it.
    // Instead, when everything is stopped for GC, we migrate all the
    // threads on the run queue then (see scheduleDoGC()).
    //
    // ToDo: what about TSO_LOCKED?  Currently we're migrating those
    // when the number of capabilities drops, but we never migrate
    // them back if it rises again.  Presumably we should, but after
    // the thread has been migrated we no longer know what capability
    // it was originally on.
#if defined(THREADED_RTS)
    if (cap->disabled && !t->bound) {
        Capability *dest_cap = capabilities[cap->no % enabled_capabilities];
        migrateThread(cap, t, dest_cap);
        continue;
    }
#endif

    /* context switches are initiated by the timer signal, unless
     * the user specified "context switch as often as possible", with
     * +RTS -C0
     */
    if (RtsFlags.ConcFlags.ctxtSwitchTicks == 0
        && !emptyThreadQueues(cap)) {
        RELAXED_STORE(&cap->context_switch, 1);
    }

run_thread:

    // CurrentTSO is the thread to run. It might be different if we
    // loop back to run_thread, so make sure to set CurrentTSO after
    // that.
    cap->r.rCurrentTSO = t;

    resumeHeapProfTimer();

    // ----------------------------------------------------------------------
    // Run the current thread

    ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
    ASSERT(t->cap == cap);
    ASSERT(t->bound ? t->bound->task->cap == cap : 1);

    prev_what_next = t->what_next;

    errno = t->saved_errno;
#if defined(mingw32_HOST_OS)
    SetLastError(t->saved_winerror);
#endif

    // reset the interrupt flag before running Haskell code
    RELAXED_STORE(&cap->interrupt, false);

    cap->in_haskell = true;
    RELAXED_STORE(&cap->idle, false);

    dirty_TSO(cap,t);
    dirty_STACK(cap,t->stackobj);

    switch (SEQ_CST_LOAD(&recent_activity))
    {
    case ACTIVITY_DONE_GC: {
        // ACTIVITY_DONE_GC means we turned off the timer signal to
        // conserve power (see #1623).  Re-enable it here.
        uint32_t prev;
        prev = xchg((P_)&recent_activity, ACTIVITY_YES);
        if (prev == ACTIVITY_DONE_GC) {
#if !defined(PROFILING)
            startTimer();
#endif
        }
        break;
    }
    case ACTIVITY_INACTIVE:
        // If we reached ACTIVITY_INACTIVE, then don't reset it until
        // we've done the GC.  The thread running here might just be
        // the IO manager thread that handle_tick() woke up via
        // wakeUpRts().
        break;
    default:
        SEQ_CST_STORE(&recent_activity, ACTIVITY_YES);
    }

    traceEventRunThread(cap, t);

    switch (prev_what_next) {

    case ThreadKilled:
    case ThreadComplete:
        /* Thread already finished, return to scheduler. */
        ret = ThreadFinished;
        break;

    case ThreadRunGHC:
    {
        StgRegTable *r;
        r = StgRun((StgFunPtr) stg_returnToStackTop, &cap->r);
        cap = regTableToCapability(r);
        ret = r->rRet;
        break;
    }

    case ThreadInterpret:
        cap = interpretBCO(cap);
        ret = cap->r.rRet;
        break;

    default:
        barf("schedule: invalid prev_what_next=%u field", prev_what_next);
    }

    cap->in_haskell = false;

    // The TSO might have moved, eg. if it re-entered the RTS and a GC
    // happened.  So find the new location:
    t = cap->r.rCurrentTSO;

    // cap->r.rCurrentTSO is charged for calls to allocate(), so we
    // don't want it set when not running a Haskell thread.
    cap->r.rCurrentTSO = NULL;

    // And save the current errno in this thread.
    // XXX: possibly bogus for SMP because this thread might already
    // be running again, see code below.
    t->saved_errno = errno;
#if defined(mingw32_HOST_OS)
    // Similarly for Windows error code
    t->saved_winerror = GetLastError();
#endif

    if (ret == ThreadBlocked) {
        if (t->why_blocked == BlockedOnBlackHole) {
            StgTSO *owner = blackHoleOwner(t->block_info.bh->bh);
            traceEventStopThread(cap, t, t->why_blocked + 6,
                                 owner != NULL ? owner->id : 0);
        } else {
            traceEventStopThread(cap, t, t->why_blocked + 6, 0);
        }
    } else {
        if (ret == StackOverflow) {
          traceEventStopThread(cap, t, ret, t->tot_stack_size);
        } else {
          traceEventStopThread(cap, t, ret, 0);
        }
    }

    ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
    ASSERT(t->cap == cap);

    // ----------------------------------------------------------------------

    // Costs for the scheduler are assigned to CCS_SYSTEM
    pauseHeapProfTimer();
#if defined(PROFILING)
    cap->r.rCCCS = CCS_SYSTEM;
#endif

    schedulePostRunThread(cap,t);

    ready_to_gc = false;

    switch (ret) {
    case HeapOverflow:
        ready_to_gc = scheduleHandleHeapOverflow(cap,t);
        break;

    case StackOverflow:
        // just adjust the stack for this thread, then pop it back
        // on the run queue.
        threadStackOverflow(cap, t);
        pushOnRunQueue(cap,t);
        break;

    case ThreadYielding:
        if (scheduleHandleYield(cap, t, prev_what_next)) {
            // shortcut for switching between compiler/interpreter:
            goto run_thread;
        }
        break;

    case ThreadBlocked:
        scheduleHandleThreadBlocked(t);
        break;

    case ThreadFinished:
        if (scheduleHandleThreadFinished(cap, task, t)) return cap;
        ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);
        break;

    default:
      barf("schedule: invalid thread return code %d", (int)ret);
    }

    if (ready_to_gc || scheduleNeedHeapProfile(ready_to_gc)) {
      scheduleDoGC(&cap,task,false,ready_to_gc,false);
    }
  } /* end of while() */
}

/* -----------------------------------------------------------------------------
 * Run queue operations
 * -------------------------------------------------------------------------- */

static void
removeFromRunQueue (Capability *cap, StgTSO *tso)
{
    if (tso->block_info.prev == END_TSO_QUEUE) {
        ASSERT(cap->run_queue_hd == tso);
        cap->run_queue_hd = tso->_link;
    } else {
        setTSOLink(cap, tso->block_info.prev, tso->_link);
    }
    if (tso->_link == END_TSO_QUEUE) {
        ASSERT(cap->run_queue_tl == tso);
        cap->run_queue_tl = tso->block_info.prev;
    } else {
        setTSOPrev(cap, tso->_link, tso->block_info.prev);
    }
    tso->_link = tso->block_info.prev = END_TSO_QUEUE;
    cap->n_run_queue--;

    IF_DEBUG(sanity, checkRunQueue(cap));
}

void
promoteInRunQueue (Capability *cap, StgTSO *tso)
{
    removeFromRunQueue(cap, tso);
    pushOnRunQueue(cap, tso);
}

/* -----------------------------------------------------------------------------
 * scheduleFindWork()
 *
 * Search for work to do, and handle messages from elsewhere.
 * -------------------------------------------------------------------------- */

static void
scheduleFindWork (Capability **pcap)
{
#if defined(mingw32_HOST_OS) && !defined(THREADED_RTS)
    queueIOThread();
#endif
    scheduleStartSignalHandlers(*pcap);

    scheduleProcessInbox(pcap);

    scheduleCheckBlockedThreads(*pcap);

#if defined(THREADED_RTS)
    if (emptyRunQueue(*pcap)) { scheduleActivateSpark(*pcap); }
#endif
}

#if defined(THREADED_RTS)
STATIC_INLINE bool
shouldYieldCapability (Capability *cap, Task *task, bool didGcLast)
{
    // we need to yield this capability to someone else if..
    //   - another thread is initiating a GC, and we didn't just do a GC
    //     (see Note [GC livelock])
    //   - another Task is returning from a foreign call
    //   - the thread at the head of the run queue cannot be run
    //     by this Task (it is bound to another Task, or it is unbound
    //     and this task it bound).
    //
    // Note [GC livelock]
    // ~~~~~~~~~~~~~~~~~~
    // If we are interrupted to do a GC, then we do not immediately do
    // another one.  This avoids a starvation situation where one
    // Capability keeps forcing a GC and the other Capabilities make no
    // progress at all.

    // Note [Data race in shouldYieldCapability]
    // ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
    // We would usually need to hold cap->lock to look at n_returning_tasks but
    // we don't here since this is just an approximate predicate. We
    // consequently need to use atomic accesses whenever touching
    // n_returning_tasks. However, since this is an approximate predicate we can
    // use a RELAXED ordering.

    return ((RELAXED_LOAD(&pending_sync) && !didGcLast) ||
            RELAXED_LOAD(&cap->n_returning_tasks) != 0 ||
            (!emptyRunQueue(cap) && (task->incall->tso == NULL
                                     ? peekRunQueue(cap)->bound != NULL
                                     : peekRunQueue(cap)->bound != task->incall)));
}

// This is the single place where a Task goes to sleep.  There are
// two reasons it might need to sleep:
//    - there are no threads to run
//    - we need to yield this Capability to someone else
//      (see shouldYieldCapability())
//
// Careful: the scheduler loop is quite delicate.  Make sure you run
// the tests in testsuite/concurrent (all ways) after modifying this,
// and also check the benchmarks in nofib/parallel for regressions.

static void
scheduleYield (Capability **pcap, Task *task)
{
    Capability *cap = *pcap;
    bool didGcLast = false;

    // if we have work, and we don't need to give up the Capability, continue.
    //
    if (!shouldYieldCapability(cap,task,false) &&
        (!emptyRunQueue(cap) ||
         !emptyInbox(cap) ||
         RELAXED_LOAD(&sched_state) >= SCHED_INTERRUPTING)) {
        return;
    }

    // otherwise yield (sleep), and keep yielding if necessary.
    do {
        if (doIdleGCWork(cap, false)) {
            // there's more idle GC work to do
            didGcLast = false;
        } else {
            // no more idle GC work to do
            didGcLast = yieldCapability(&cap,task, !didGcLast);
        }
    }
    while (shouldYieldCapability(cap,task,didGcLast));

    // note there may still be no threads on the run queue at this
    // point, the caller has to check.

    *pcap = cap;
    return;
}
#endif

/* -----------------------------------------------------------------------------
 * schedulePushWork()
 *
 * Push work to other Capabilities if we have some.
 * -------------------------------------------------------------------------- */

static void
schedulePushWork(Capability *cap USED_IF_THREADS,
                 Task *task      USED_IF_THREADS)
{
#if defined(THREADED_RTS)

    Capability *free_caps[n_capabilities], *cap0;
    uint32_t i, n_wanted_caps, n_free_caps;

    uint32_t spare_threads = cap->n_run_queue > 0 ? cap->n_run_queue - 1 : 0;

    // migration can be turned off with +RTS -qm
    if (!RtsFlags.ParFlags.migrate) {
        spare_threads = 0;
    }

    // Figure out how many capabilities we want to wake up.  We need at least
    // sparkPoolSize(cap) plus the number of spare threads we have.
    n_wanted_caps = sparkPoolSizeCap(cap) + spare_threads;
    if (n_wanted_caps == 0) return;

    // First grab as many free Capabilities as we can.  ToDo: we should use
    // capabilities on the same NUMA node preferably, but not exclusively.
    for (i = (cap->no + 1) % n_capabilities, n_free_caps=0;
         n_free_caps < n_wanted_caps && i != cap->no;
         i = (i + 1) % n_capabilities) {
        cap0 = capabilities[i];
        if (cap != cap0 && !cap0->disabled && tryGrabCapability(cap0,task)) {
            if (!emptyRunQueue(cap0)
                || RELAXED_LOAD(&cap0->n_returning_tasks) != 0
                || !emptyInbox(cap0)) {
                // it already has some work, we just grabbed it at
                // the wrong moment.  Or maybe it's deadlocked!
                releaseCapability(cap0);
            } else {
                free_caps[n_free_caps++] = cap0;
            }
        }
    }

    // We now have n_free_caps free capabilities stashed in
    // free_caps[].  Attempt to share our run queue equally with them.
    // This is complicated slightly by the fact that we can't move
    // some threads:
    //
    //  - threads that have TSO_LOCKED cannot migrate
    //  - a thread that is bound to the current Task cannot be migrated
    //
    // This is about the simplest thing we could do; improvements we
    // might want to do include:
    //
    //   - giving high priority to moving relatively new threads, on
    //     the gournds that they haven't had time to build up a
    //     working set in the cache on this CPU/Capability.
    //
    //   - giving low priority to moving long-lived threads

    if (n_free_caps > 0) {
        StgTSO *prev, *t, *next;

        debugTrace(DEBUG_sched,
                   "cap %d: %d threads, %d sparks, and %d free capabilities, sharing...",
                   cap->no, cap->n_run_queue, sparkPoolSizeCap(cap),
                   n_free_caps);

        // There are n_free_caps+1 caps in total.  We will share the threads
        // evently between them, *except* that if the run queue does not divide
        // evenly by n_free_caps+1 then we bias towards the current capability.
        // e.g. with n_run_queue=4, n_free_caps=2, we will keep 2.
        uint32_t keep_threads =
            (cap->n_run_queue + n_free_caps) / (n_free_caps + 1);

        // This also ensures that we don't give away all our threads, since
        // (x + y) / (y + 1) >= 1 when x >= 1.

        // The number of threads we have left.
        uint32_t n = cap->n_run_queue;

        // prev = the previous thread on this cap's run queue
        prev = END_TSO_QUEUE;

        // We're going to walk through the run queue, migrating threads to other
        // capabilities until we have only keep_threads left.  We might
        // encounter a thread that cannot be migrated, in which case we add it
        // to the current run queue and decrement keep_threads.
        for (t = cap->run_queue_hd, i = 0;
             t != END_TSO_QUEUE && n > keep_threads;
             t = next)
        {
            next = t->_link;
            t->_link = END_TSO_QUEUE;

            // Should we keep this thread?
            if (t->bound == task->incall // don't move my bound thread
                || tsoLocked(t) // don't move a locked thread
                ) {
                if (prev == END_TSO_QUEUE) {
                    cap->run_queue_hd = t;
                } else {
                    setTSOLink(cap, prev, t);
                }
                setTSOPrev(cap, t, prev);
                prev = t;
                if (keep_threads > 0) keep_threads--;
            }

            // Or migrate it?
            else {
                appendToRunQueue(free_caps[i],t);
                traceEventMigrateThread (cap, t, free_caps[i]->no);

                // See Note [Benign data race due to work-pushing].
                if (t->bound) {
                    t->bound->task->cap = free_caps[i];
                }
                t->cap = free_caps[i];
                n--; // we have one fewer threads now
                i++; // move on to the next free_cap
                if (i == n_free_caps) i = 0;
            }
        }

        // Join up the beginning of the queue (prev)
        // with the rest of the queue (t)
        if (t == END_TSO_QUEUE) {
            cap->run_queue_tl = prev;
        } else {
            setTSOPrev(cap, t, prev);
        }
        if (prev == END_TSO_QUEUE) {
            cap->run_queue_hd = t;
        } else {
            setTSOLink(cap, prev, t);
        }
        cap->n_run_queue = n;

        IF_DEBUG(sanity, checkRunQueue(cap));

        // release the capabilities
        for (i = 0; i < n_free_caps; i++) {
            task->cap = free_caps[i];
            if (sparkPoolSizeCap(cap) > 0) {
                // If we have sparks to steal, wake up a worker on the
                // capability, even if it has no threads to run.
                releaseAndWakeupCapability(free_caps[i]);
            } else {
                releaseCapability(free_caps[i]);
            }
        }
    }
    task->cap = cap; // reset to point to our Capability.

#endif /* THREADED_RTS */

}

/* ----------------------------------------------------------------------------
 * Start any pending signal handlers
 * ------------------------------------------------------------------------- */

#if defined(RTS_USER_SIGNALS) && !defined(THREADED_RTS)
static void
scheduleStartSignalHandlers(Capability *cap)
{
    if (RtsFlags.MiscFlags.install_signal_handlers && signals_pending()) {
        // safe outside the lock
        startSignalHandlers(cap);
    }
}
#else
static void
scheduleStartSignalHandlers(Capability *cap STG_UNUSED)
{
}
#endif

/* ----------------------------------------------------------------------------
 * Check for blocked threads that can be woken up.
 * ------------------------------------------------------------------------- */

static void
scheduleCheckBlockedThreads(Capability *cap USED_IF_NOT_THREADS)
{
#if !defined(THREADED_RTS)
    //
    // Check whether any waiting threads need to be woken up.  If the
    // run queue is empty, and there are no other tasks running, we
    // can wait indefinitely for something to happen.
    //
    if ( !emptyQueue(blocked_queue_hd) || !emptyQueue(sleeping_queue) )
    {
        awaitEvent (emptyRunQueue(cap));
    }
#endif
}

/* ----------------------------------------------------------------------------
 * Detect deadlock conditions and attempt to resolve them.
 * ------------------------------------------------------------------------- */

static void
scheduleDetectDeadlock (Capability **pcap, Task *task)
{
    Capability *cap = *pcap;
    /*
     * Detect deadlock: when we have no threads to run, there are no
     * threads blocked, waiting for I/O, or sleeping, and all the
     * other tasks are waiting for work, we must have a deadlock of
     * some description.
     */
    if ( emptyThreadQueues(cap) )
    {
#if defined(THREADED_RTS)
        /*
         * In the threaded RTS, we only check for deadlock if there
         * has been no activity in a complete timeslice.  This means
         * we won't eagerly start a full GC just because we don't have
         * any threads to run currently.
         */
        if (SEQ_CST_LOAD(&recent_activity) != ACTIVITY_INACTIVE) return;
#endif

        debugTrace(DEBUG_sched, "deadlocked, forcing major GC...");

        // Garbage collection can release some new threads due to
        // either (a) finalizers or (b) threads resurrected because
        // they are unreachable and will therefore be sent an
        // exception.  Any threads thus released will be immediately
        // runnable.
        scheduleDoGC (pcap, task, true/*force major GC*/, false /* Whether it is an overflow GC */, true/*deadlock detection*/);
        cap = *pcap;
        // when force_major == true. scheduleDoGC sets
        // recent_activity to ACTIVITY_DONE_GC and turns off the timer
        // signal.

        if ( !emptyRunQueue(cap) ) return;

#if defined(RTS_USER_SIGNALS) && !defined(THREADED_RTS)
        /* If we have user-installed signal handlers, then wait
         * for signals to arrive rather then bombing out with a
         * deadlock.
         */
        if ( RtsFlags.MiscFlags.install_signal_handlers && anyUserHandlers() ) {
            debugTrace(DEBUG_sched,
                       "still deadlocked, waiting for signals...");

            awaitUserSignals();

            if (signals_pending()) {
                startSignalHandlers(cap);
            }

            // either we have threads to run, or we were interrupted:
            ASSERT(!emptyRunQueue(cap) || sched_state >= SCHED_INTERRUPTING);

            return;
        }
#endif
    }
}


/* ----------------------------------------------------------------------------
 * Process message in the current Capability's inbox
 * ------------------------------------------------------------------------- */

static void
scheduleProcessInbox (Capability **pcap USED_IF_THREADS)
{
#if defined(THREADED_RTS)
    Message *m, *next;
    PutMVar *p, *pnext;
    int r;
    Capability *cap = *pcap;

    while (!emptyInbox(cap)) {
        // Executing messages might use heap, so we should check for GC.
        if (doYouWantToGC(cap)) {
            scheduleDoGC(pcap, cap->running_task, false, false, false);
            cap = *pcap;
        }

        // don't use a blocking acquire; if the lock is held by
        // another thread then just carry on.  This seems to avoid
        // getting stuck in a message ping-pong situation with other
        // processors.  We'll check the inbox again later anyway.
        //
        // We should really use a more efficient queue data structure
        // here.  The trickiness is that we must ensure a Capability
        // never goes idle if the inbox is non-empty, which is why we
        // use cap->lock (cap->lock is released as the last thing
        // before going idle; see Capability.c:releaseCapability()).
        r = TRY_ACQUIRE_LOCK(&cap->lock);
        if (r != 0) return;

        m = cap->inbox;
        p = cap->putMVars;
        cap->inbox = (Message*)END_TSO_QUEUE;
        cap->putMVars = NULL;

        RELEASE_LOCK(&cap->lock);

        while (m != (Message*)END_TSO_QUEUE) {
            next = m->link;
            executeMessage(cap, m);
            m = next;
        }

        while (p != NULL) {
            pnext = p->link;
            performTryPutMVar(cap, (StgMVar*)deRefStablePtr(p->mvar),
                              Unit_closure);
            freeStablePtr(p->mvar);
            stgFree(p);
            p = pnext;
        }
    }
#endif
}


/* ----------------------------------------------------------------------------
 * Activate spark threads (THREADED_RTS)
 * ------------------------------------------------------------------------- */

#if defined(THREADED_RTS)
static void
scheduleActivateSpark(Capability *cap)
{
    if (anySparks() && !cap->disabled)
    {
        createSparkThread(cap);
        debugTrace(DEBUG_sched, "creating a spark thread");
    }
}
#endif // THREADED_RTS

/* ----------------------------------------------------------------------------
 * After running a thread...
 * ------------------------------------------------------------------------- */

static void
schedulePostRunThread (Capability *cap, StgTSO *t)
{
    // We have to be able to catch transactions that are in an
    // infinite loop as a result of seeing an inconsistent view of
    // memory, e.g.
    //
    //   atomically $ do
    //       [a,b] <- mapM readTVar [ta,tb]
    //       when (a == b) loop
    //
    // and a is never equal to b given a consistent view of memory.
    //
    if (t -> trec != NO_TREC && t -> why_blocked == NotBlocked) {
        if (!stmValidateNestOfTransactions(cap, t -> trec)) {
            debugTrace(DEBUG_sched | DEBUG_stm,
                       "trec %p found wasting its time", t);

            // strip the stack back to the
            // ATOMICALLY_FRAME, aborting the (nested)
            // transaction, and saving the stack of any
            // partially-evaluated thunks on the heap.
            throwToSingleThreaded_(cap, t, NULL, true);

//            ASSERT(get_itbl((StgClosure *)t->sp)->type == ATOMICALLY_FRAME);
        }
    }

    //
    // If the current thread's allocation limit has run out, send it
    // the AllocationLimitExceeded exception.

    if (PK_Int64((W_*)&(t->alloc_limit)) < 0 && (t->flags & TSO_ALLOC_LIMIT)) {
        // Use a throwToSelf rather than a throwToSingleThreaded, because
        // it correctly handles the case where the thread is currently
        // inside mask.  Also the thread might be blocked (e.g. on an
        // MVar), and throwToSingleThreaded doesn't unblock it
        // correctly in that case.
        throwToSelf(cap, t, allocationLimitExceeded_closure);
        ASSIGN_Int64((W_*)&(t->alloc_limit),
                     (StgInt64)RtsFlags.GcFlags.allocLimitGrace * BLOCK_SIZE);
    }

  /* some statistics gathering in the parallel case */
}

/* -----------------------------------------------------------------------------
 * Handle a thread that returned to the scheduler with ThreadHeapOverflow
 * -------------------------------------------------------------------------- */

static bool
scheduleHandleHeapOverflow( Capability *cap, StgTSO *t )
{
    if (cap->r.rHpLim == NULL || RELAXED_LOAD(&cap->context_switch)) {
        // Sometimes we miss a context switch, e.g. when calling
        // primitives in a tight loop, MAYBE_GC() doesn't check the
        // context switch flag, and we end up waiting for a GC.
        // See #1984, and concurrent/should_run/1984
        RELAXED_STORE(&cap->context_switch, 0);
        appendToRunQueue(cap,t);
    } else {
        pushOnRunQueue(cap,t);
    }

    // did the task ask for a large block?
    if (cap->r.rHpAlloc > BLOCK_SIZE) {
        // if so, get one and push it on the front of the nursery.
        bdescr *bd;
        W_ blocks;

        blocks = (W_)BLOCK_ROUND_UP(cap->r.rHpAlloc) / BLOCK_SIZE;

        if (blocks > BLOCKS_PER_MBLOCK) {
            barf("allocation of %ld bytes too large (GHC should have complained at compile-time)", (long)cap->r.rHpAlloc);
        }

        debugTrace(DEBUG_sched,
                   "--<< thread %ld (%s) stopped: requesting a large block (size %ld)\n",
                   (long)t->id, what_next_strs[t->what_next], blocks);

        // don't do this if the nursery is (nearly) full, we'll GC first.
        if (cap->r.rCurrentNursery->link != NULL ||
            cap->r.rNursery->n_blocks == 1) {  // paranoia to prevent
                                               // infinite loop if the
                                               // nursery has only one
                                               // block.

            bd = allocGroupOnNode_lock(cap->node,blocks);
            cap->r.rNursery->n_blocks += blocks;

            // link the new group after CurrentNursery
            dbl_link_insert_after(bd, cap->r.rCurrentNursery);

            // initialise it as a nursery block.  We initialise the
            // step, gen_no, and flags field of *every* sub-block in
            // this large block, because this is easier than making
            // sure that we always find the block head of a large
            // block whenever we call Bdescr() (eg. evacuate() and
            // isAlive() in the GC would both have to do this, at
            // least).
            {
                bdescr *x;
                for (x = bd; x < bd + blocks; x++) {
                    initBdescr(x,g0,g0);
                    x->free = x->start;
                    x->flags = 0;
                }
            }

            // This assert can be a killer if the app is doing lots
            // of large block allocations.
            IF_DEBUG(sanity, checkNurserySanity(cap->r.rNursery));

            // now update the nursery to point to the new block
            finishedNurseryBlock(cap, cap->r.rCurrentNursery);
            cap->r.rCurrentNursery = bd;

            // we might be unlucky and have another thread get on the
            // run queue before us and steal the large block, but in that
            // case the thread will just end up requesting another large
            // block.
            return false;  /* not actually GC'ing */
        }
    }

    return doYouWantToGC(cap);
    /* actual GC is done at the end of the while loop in schedule() */
}

/* -----------------------------------------------------------------------------
 * Handle a thread that returned to the scheduler with ThreadYielding
 * -------------------------------------------------------------------------- */

static bool
scheduleHandleYield( Capability *cap, StgTSO *t, uint32_t prev_what_next )
{
    /* put the thread back on the run queue.  Then, if we're ready to
     * GC, check whether this is the last task to stop.  If so, wake
     * up the GC thread.  getThread will block during a GC until the
     * GC is finished.
     */

    ASSERT(t->_link == END_TSO_QUEUE);

    // Shortcut if we're just switching evaluators: just run the thread.  See
    // Note [avoiding threadPaused] in Interpreter.c.
    if (t->what_next != prev_what_next) {
        debugTrace(DEBUG_sched,
                   "--<< thread %ld (%s) stopped to switch evaluators",
                   (long)t->id, what_next_strs[t->what_next]);
        return true;
    }

    // Reset the context switch flag.  We don't do this just before
    // running the thread, because that would mean we would lose ticks
    // during GC, which can lead to unfair scheduling (a thread hogs
    // the CPU because the tick always arrives during GC).  This way
    // penalises threads that do a lot of allocation, but that seems
    // better than the alternative.
    if (RELAXED_LOAD(&cap->context_switch) != 0) {
        RELAXED_STORE(&cap->context_switch, 0);
        appendToRunQueue(cap,t);
    } else {
        pushOnRunQueue(cap,t);
    }

    IF_DEBUG(sanity,
             //debugBelch("&& Doing sanity check on yielding TSO %ld.", t->id);
             checkTSO(t));

    return false;
}

/* -----------------------------------------------------------------------------
 * Handle a thread that returned to the scheduler with ThreadBlocked
 * -------------------------------------------------------------------------- */

static void
scheduleHandleThreadBlocked( StgTSO *t
#if !defined(DEBUG)
    STG_UNUSED
#endif
    )
{

      // We don't need to do anything.  The thread is blocked, and it
      // has tidied up its stack and placed itself on whatever queue
      // it needs to be on.

    // ASSERT(t->why_blocked != NotBlocked);
    // Not true: for example,
    //    - the thread may have woken itself up already, because
    //      threadPaused() might have raised a blocked throwTo
    //      exception, see maybePerformBlockedException().

#if defined(DEBUG)
    traceThreadStatus(DEBUG_sched, t);
#endif
}

/* -----------------------------------------------------------------------------
 * Handle a thread that returned to the scheduler with ThreadFinished
 * -------------------------------------------------------------------------- */

static bool
scheduleHandleThreadFinished (Capability *cap, Task *task, StgTSO *t)
{
    /* Need to check whether this was a main thread, and if so,
     * return with the return value.
     *
     * We also end up here if the thread kills itself with an
     * uncaught exception, see Exception.cmm.
     */

    // blocked exceptions can now complete, even if the thread was in
    // blocked mode (see #2910).
    awakenBlockedExceptionQueue (cap, t);

      //
      // Check whether the thread that just completed was a bound
      // thread, and if so return with the result.
      //
      // There is an assumption here that all thread completion goes
      // through this point; we need to make sure that if a thread
      // ends up in the ThreadKilled state, that it stays on the run
      // queue so it can be dealt with here.
      //

      if (t->bound) {

          if (t->bound != task->incall) {
#if !defined(THREADED_RTS)
              // Must be a bound thread that is not the topmost one.  Leave
              // it on the run queue until the stack has unwound to the
              // point where we can deal with this.  Leaving it on the run
              // queue also ensures that the garbage collector knows about
              // this thread and its return value (it gets dropped from the
              // step->threads list so there's no other way to find it).
              appendToRunQueue(cap,t);
              return false;
#else
              // this cannot happen in the threaded RTS, because a
              // bound thread can only be run by the appropriate Task.
              barf("finished bound thread that isn't mine");
#endif
          }

          ASSERT(task->incall->tso == t);

          if (t->what_next == ThreadComplete) {
              if (task->incall->ret) {
                  // NOTE: return val is stack->sp[1] (see StgStartup.cmm)
                  StgDeadThreadFrame *dead = (StgDeadThreadFrame *) &task->incall->tso->stackobj->sp[0];
                  ASSERT(dead->header.info == &stg_dead_thread_info);
                  *(task->incall->ret) = (StgClosure *) dead->result;
              }
              task->incall->rstat = Success;
          } else {
              if (task->incall->ret) {
                  *(task->incall->ret) = NULL;
              }
              if (RELAXED_LOAD(&sched_state) >= SCHED_INTERRUPTING) {
                  if (heap_overflow) {
                      task->incall->rstat = HeapExhausted;
                  } else {
                      task->incall->rstat = Interrupted;
                  }
              } else {
                  task->incall->rstat = Killed;
              }
          }

          removeThreadLabel(task->incall->tso->id);

          // We no longer consider this thread and task to be bound to
          // each other.  The TSO lives on until it is GC'd, but the
          // task is about to be released by the caller, and we don't
          // want anyone following the pointer from the TSO to the
          // defunct task (which might have already been
          // re-used). This was a real bug: the GC updated
          // tso->bound->tso which lead to a deadlock.
          t->bound = NULL;
          task->incall->tso = NULL;

          return true; // tells schedule() to return
      }

      return false;
}

/* -----------------------------------------------------------------------------
 * Perform a heap census
 * -------------------------------------------------------------------------- */

static bool
scheduleNeedHeapProfile( bool ready_to_gc )
{
    // When we have +RTS -i0 and we're heap profiling, do a census at
    // every GC.  This lets us get repeatable runs for debugging.
    if (performHeapProfile ||
        (RtsFlags.ProfFlags.heapProfileInterval==0 &&
         RtsFlags.ProfFlags.doHeapProfile && ready_to_gc)) {
        return true;
    } else {
        return false;
    }
}

/* -----------------------------------------------------------------------------
 * stopAllCapabilities()
 *
 * Stop all Haskell execution.  This is used when we need to make some global
 * change to the system, such as altering the number of capabilities, or
 * forking.
 *
 * pCap may be NULL in the event that the caller doesn't yet own a capability.
 *
 * To resume after stopAllCapabilities(), use releaseAllCapabilities().
 * -------------------------------------------------------------------------- */

#if defined(THREADED_RTS)
void stopAllCapabilities
    ( Capability **pCap     // [in/out] This thread's task's owned capability.
                            //          pCap may be NULL if no capability is owned.
                            //          Else *pCap != NULL
                            // On return, set to the task's newly owned
                            // capability (task->cap). Though, the Task will
                            // technically own all capabilities.
    , Task *task            // [in] This thread's task.
    )
{
    stopAllCapabilitiesWith(pCap, task, SYNC_OTHER);
}

void stopAllCapabilitiesWith (Capability **pCap, Task *task, SyncType sync_type)
{
    bool was_syncing;
    SyncType prev_sync_type;

    PendingSync sync = {
        .type = sync_type,
        .idle = NULL,
        .task = task
    };

    do {
        was_syncing = requestSync(pCap, task, &sync, &prev_sync_type);
    } while (was_syncing);

    acquireAllCapabilities(pCap ? *pCap : NULL, task);

    pending_sync = 0;
    signalCondition(&sync_finished_cond);
}
#endif

/* -----------------------------------------------------------------------------
 * requestSync()
 *
 * Commence a synchronisation between all capabilities.  Normally not called
 * directly, instead use stopAllCapabilities().  This is used by the GC, which
 * has some special synchronisation requirements.
 *
 * Note that this can be called in two ways:
 *
 * - where *pcap points to a capability owned by the caller: in this case
 *   *prev_sync_type will reflect the in-progress sync type on return, if one
 *   *was found
 *
 *  - where pcap == NULL: in this case the caller doesn't hold a capability.
 *    we only return whether or not a pending sync was found and prev_sync_type
 *    is unchanged.
 *
 * Returns:
 *    false if we successfully got a sync
 *    true  if there was another sync request in progress,
 *             and we yielded to it.  The value returned is the
 *             type of the other sync request.
 * -------------------------------------------------------------------------- */

#if defined(THREADED_RTS)
static bool requestSync
    ( Capability **pcap         // [in/out] This thread's task's owned capability.
                                // May change if there is an existing sync (true is returned).
                                // Precondition:
                                //      pcap may be NULL
                                //      *pcap != NULL
    , Task *task                // [in] This thread's task.
    , PendingSync *new_sync     // [in] The new requested sync.
    , SyncType *prev_sync_type  // [out] Only set if there is an existing sync (true is returned).
    )
{
    PendingSync *sync;

    sync = (PendingSync*)cas((StgVolatilePtr)&pending_sync,
                             (StgWord)NULL,
                             (StgWord)new_sync);

    if (sync != NULL)
    {
        // sync is valid until we have called yieldCapability().
        // After the sync is completed, we cannot read that struct any
        // more because it has been freed.
        *prev_sync_type = sync->type;
        if (pcap == NULL) {
            // The caller does not hold a capability (e.g. may be a concurrent
            // mark thread). Consequently we must wait until the pending sync is
            // finished before proceeding to ensure we don't loop.
            // TODO: Don't busy-wait
            ACQUIRE_LOCK(&sync_finished_mutex);
            while (pending_sync) {
                waitCondition(&sync_finished_cond, &sync_finished_mutex);
            }
            RELEASE_LOCK(&sync_finished_mutex);
        } else {
            do {
                debugTrace(DEBUG_sched, "someone else is trying to sync (%d)...",
                          sync->type);
                ASSERT(*pcap);
                yieldCapability(pcap,task,true);
                sync = SEQ_CST_LOAD(&pending_sync);
            } while (sync != NULL);
        }

        // NOTE: task->cap might have changed now
        return true;
    }
    else
    {
        return false;
    }
}
#endif

/* -----------------------------------------------------------------------------
 * acquireAllCapabilities()
 *
 * Grab all the capabilities except the one we already hold (cap may be NULL is
 * the caller does not currently hold a capability). Used when synchronising
 * before a single-threaded GC (SYNC_SEQ_GC), and before a fork (SYNC_OTHER).
 *
 * Only call this after requestSync(), otherwise a deadlock might
 * ensue if another thread is trying to synchronise.
 * -------------------------------------------------------------------------- */

#if defined(THREADED_RTS)
static void acquireAllCapabilities(Capability *cap, Task *task)
{
    Capability *tmpcap;
    uint32_t i;

    ASSERT(SEQ_CST_LOAD(&pending_sync) != NULL);
    for (i=0; i < n_capabilities; i++) {
        debugTrace(DEBUG_sched, "grabbing all the capabilies (%d/%d)",
                   i, n_capabilities);
        tmpcap = capabilities[i];
        if (tmpcap != cap) {
            // we better hope this task doesn't get migrated to
            // another Capability while we're waiting for this one.
            // It won't, because load balancing happens while we have
            // all the Capabilities, but even so it's a slightly
            // unsavoury invariant.
            task->cap = tmpcap;
            waitForCapability(&tmpcap, task);
            if (tmpcap->no != i) {
                barf("acquireAllCapabilities: got the wrong capability");
            }
        }
    }
    task->cap = cap == NULL ? tmpcap : cap;
}
#endif

/* -----------------------------------------------------------------------------
 * releaseAllCapabilities()
 *
 * Assuming this thread holds all the capabilities, release them all (except for
 * the one passed in as keep_cap, if non-NULL).
 * -------------------------------------------------------------------------- */

#if defined(THREADED_RTS)
void releaseAllCapabilities(uint32_t n, Capability *keep_cap, Task *task)
{
    uint32_t i;
    ASSERT( task != NULL);
    for (i = 0; i < n; i++) {
        Capability *tmpcap = capabilities[i];
        if (keep_cap != tmpcap) {
            task->cap = tmpcap;
            releaseCapability(tmpcap);
        }
    }
    task->cap = keep_cap;
}
#endif

/* -----------------------------------------------------------------------------
 * Perform a garbage collection if necessary
 * -------------------------------------------------------------------------- */

// N.B. See Note [Deadlock detection under nonmoving collector] for rationale
// behind deadlock_detect argument.
static void
scheduleDoGC (Capability **pcap, Task *task USED_IF_THREADS,
              bool force_major, bool is_overflow_gc, bool deadlock_detect)
{
    Capability *cap = *pcap;
    bool heap_census;
    uint32_t collect_gen;
    bool major_gc;
#if defined(THREADED_RTS)
    uint32_t gc_type;
    uint32_t i;
    uint32_t need_idle;
    uint32_t n_gc_threads;
    uint32_t n_idle_caps = 0, n_failed_trygrab_idles = 0;
    StgTSO *tso;
    bool *idle_cap;
      // idle_cap is an array (allocated later) of size n_capabilities, where
      // idle_cap[i] is rtsTrue if capability i will be idle during this GC
      // cycle.
#endif

    if (RELAXED_LOAD(&sched_state) == SCHED_SHUTTING_DOWN) {
        // The final GC has already been done, and the system is
        // shutting down.  We'll probably deadlock if we try to GC
        // now.
        return;
    }

    heap_census = scheduleNeedHeapProfile(true);

    // Figure out which generation we are collecting, so that we can
    // decide whether this is a parallel GC or not.
    collect_gen = calcNeeded(force_major || heap_census, NULL);
    major_gc = (collect_gen == RtsFlags.GcFlags.generations-1);

#if defined(THREADED_RTS)
    if (RELAXED_LOAD(&sched_state) < SCHED_INTERRUPTING
        && RtsFlags.ParFlags.parGcEnabled
        && collect_gen >= RtsFlags.ParFlags.parGcGen
        && ! oldest_gen->mark)
    {
        gc_type = SYNC_GC_PAR;
    } else {
        gc_type = SYNC_GC_SEQ;
    }

    // In order to GC, there must be no threads running Haskell code.
    // Therefore, for single-threaded GC, the GC thread needs to hold *all* the
    // capabilities, and release them after the GC has completed.  For parallel
    // GC, we synchronise all the running threads using requestSync().
    //
    // Other capabilities are prevented from running yet more Haskell threads if
    // pending_sync is set. Tested inside yieldCapability() and
    // releaseCapability() in Capability.c

    PendingSync sync = {
        .type = gc_type,
        .idle = NULL,
        .task = task
    };

    {
        SyncType prev_sync = 0;
        bool was_syncing;
        do {
            // If -qn is not set and we have more capabilities than cores, set
            // the number of GC threads to #cores.  We do this here rather than
            // in normaliseRtsOpts() because here it will work if the program
            // calls setNumCapabilities.
            //
            n_gc_threads = RtsFlags.ParFlags.parGcThreads;
            if (n_gc_threads == 0 &&
                enabled_capabilities > getNumberOfProcessors()) {
                n_gc_threads = getNumberOfProcessors();
            }

            // This calculation must be inside the loop because
            // enabled_capabilities may change if requestSync() below fails and
            // we retry.
            if (gc_type == SYNC_GC_PAR && n_gc_threads > 0) {
                if (n_gc_threads >= enabled_capabilities) {
                    need_idle = 0;
                } else {
                    need_idle = enabled_capabilities - n_gc_threads;
                }
            } else {
                need_idle = 0;
            }

            // We need an array of size n_capabilities, but since this may
            // change each time around the loop we must allocate it afresh.
            idle_cap = (bool *)stgMallocBytes(n_capabilities *
                                              sizeof(bool),
                                              "scheduleDoGC");
            sync.idle = idle_cap;

            // When using +RTS -qn, we need some capabilities to be idle during
            // GC.  The best bet is to choose some inactive ones, so we look for
            // those first:
            uint32_t n_idle = need_idle;
            for (i=0; i < n_capabilities; i++) {
                if (capabilities[i]->disabled) {
                    idle_cap[i] = true;
                } else if (n_idle > 0 &&
                           capabilities[i]->running_task == NULL) {
                    debugTrace(DEBUG_sched, "asking for cap %d to be idle", i);
                    n_idle--;
                    idle_cap[i] = true;
                } else {
                    idle_cap[i] = false;
                }
            }
            // If we didn't find enough inactive capabilities, just pick some
            // more to be idle.
            for (i=0; n_idle > 0 && i < n_capabilities; i++) {
                if (!idle_cap[i] && i != cap->no) {
                    idle_cap[i] = true;
                    n_idle--;
                }
            }
            ASSERT(n_idle == 0);

            was_syncing = requestSync(pcap, task, &sync, &prev_sync);
            cap = *pcap;
            if (was_syncing) {
                stgFree(idle_cap);
            }
            if (was_syncing &&
                (prev_sync == SYNC_GC_SEQ || prev_sync == SYNC_GC_PAR) &&
                !(RELAXED_LOAD(&sched_state) == SCHED_INTERRUPTING && force_major)) {
                // someone else had a pending sync request for a GC, so
                // let's assume GC has been done and we don't need to GC
                // again.
                // Exception to this: if SCHED_INTERRUPTING, then we still
                // need to do the final GC.
                return;
            }
            if (RELAXED_LOAD(&sched_state) == SCHED_SHUTTING_DOWN) {
                // The scheduler might now be shutting down.  We tested
                // this above, but it might have become true since then as
                // we yielded the capability in requestSync().
                return;
            }
        } while (was_syncing);
    }

    stat_startGCSync(gc_threads[cap->no]);

    unsigned int old_n_capabilities = n_capabilities;

    interruptAllCapabilities();

    // The final shutdown GC is always single-threaded, because it's
    // possible that some of the Capabilities have no worker threads.

    if (gc_type == SYNC_GC_SEQ) {
        traceEventRequestSeqGc(cap);
    } else {
        traceEventRequestParGc(cap);
    }

    if (gc_type == SYNC_GC_SEQ) {
        // single-threaded GC: grab all the capabilities
        acquireAllCapabilities(cap,task);
    }
    else
    {
        // If we are load-balancing collections in this
        // generation, then we require all GC threads to participate
        // in the collection.  Otherwise, we only require active
        // threads to participate, and we set gc_threads[i]->idle for
        // any idle capabilities.  The rationale here is that waking
        // up an idle Capability takes much longer than just doing any
        // GC work on its behalf.

        if (RtsFlags.ParFlags.parGcNoSyncWithIdle == 0
            || (RtsFlags.ParFlags.parGcLoadBalancingEnabled &&
                collect_gen >= RtsFlags.ParFlags.parGcLoadBalancingGen))
        {
            for (i=0; i < n_capabilities; i++) {
                if (capabilities[i]->disabled) {
                    idle_cap[i] = tryGrabCapability(capabilities[i], task);
                    if (idle_cap[i]) {
                        n_idle_caps++;
                    }
                } else {
                    if (i != cap->no && idle_cap[i]) {
                        Capability *tmpcap = capabilities[i];
                        task->cap = tmpcap;
                        waitForCapability(&tmpcap, task);
                        n_idle_caps++;
                    }
                }
            }
        }
        else
        {
            for (i=0; i < n_capabilities; i++) {
                if (capabilities[i]->disabled) {
                    idle_cap[i] = tryGrabCapability(capabilities[i], task);
                    if (idle_cap[i]) {
                        n_idle_caps++;
                    }
                } else if (i != cap->no &&
                           capabilities[i]->idle >=
                           RtsFlags.ParFlags.parGcNoSyncWithIdle) {
                    idle_cap[i] = tryGrabCapability(capabilities[i], task);
                    if (idle_cap[i]) {
                        n_idle_caps++;
                    } else {
                        n_failed_trygrab_idles++;
                    }
                }
            }
        }
        debugTrace(DEBUG_sched, "%d idle caps", n_idle_caps);

        for (i=0; i < n_capabilities; i++) {
            NONATOMIC_ADD(&capabilities[i]->idle, 1);
        }

        // For all capabilities participating in this GC, wait until
        // they have stopped mutating and are standing by for GC.
        waitForGcThreads(cap, idle_cap);

        // Stable point where we can do a global check on our spark counters
        ASSERT(checkSparkCountInvariant());
    }

#endif

    IF_DEBUG(scheduler, printAllThreads());

delete_threads_and_gc:
    /*
     * We now have all the capabilities; if we're in an interrupting
     * state, then we should take the opportunity to delete all the
     * threads in the system.
     * Checking for major_gc ensures that the last GC is major.
     */
    if (RELAXED_LOAD(&sched_state) == SCHED_INTERRUPTING && major_gc) {
        deleteAllThreads();
#if defined(THREADED_RTS)
        // Discard all the sparks from every Capability.  Why?
        // They'll probably be GC'd anyway since we've killed all the
        // threads.  It just avoids the GC having to do any work to
        // figure out that any remaining sparks are garbage.
        for (i = 0; i < n_capabilities; i++) {
            capabilities[i]->spark_stats.gcd +=
                sparkPoolSize(capabilities[i]->sparks);
            // No race here since all Caps are stopped.
            discardSparksCap(capabilities[i]);
        }
#endif
        RELAXED_STORE(&sched_state, SCHED_SHUTTING_DOWN);
    }

    /*
     * When there are disabled capabilities, we want to migrate any
     * threads away from them.  Normally this happens in the
     * scheduler's loop, but only for unbound threads - it's really
     * hard for a bound thread to migrate itself.  So we have another
     * go here.
     */
#if defined(THREADED_RTS)
    for (i = enabled_capabilities; i < n_capabilities; i++) {
        Capability *tmp_cap, *dest_cap;
        tmp_cap = capabilities[i];
        ASSERT(tmp_cap->disabled);
        if (i != cap->no) {
            dest_cap = capabilities[i % enabled_capabilities];
            while (!emptyRunQueue(tmp_cap)) {
                tso = popRunQueue(tmp_cap);
                migrateThread(tmp_cap, tso, dest_cap);
                if (tso->bound) {
                  traceTaskMigrate(tso->bound->task,
                                   tso->bound->task->cap,
                                   dest_cap);
                  tso->bound->task->cap = dest_cap;
                }
            }
        }
    }
#endif

    // Do any remaining idle GC work from the previous GC
    doIdleGCWork(cap, true /* all of it */);

#if defined(THREADED_RTS)
    // reset pending_sync *before* GC, so that when the GC threads
    // emerge they don't immediately re-enter the GC.
    pending_sync = 0;
    signalCondition(&sync_finished_cond);
    GarbageCollect(collect_gen, heap_census, is_overflow_gc, deadlock_detect, gc_type, cap, idle_cap);
#else
    GarbageCollect(collect_gen, heap_census, is_overflow_gc, deadlock_detect, 0, cap, NULL);
#endif

    // If we're shutting down, don't leave any idle GC work to do.
    if (RELAXED_LOAD(&sched_state) == SCHED_SHUTTING_DOWN) {
        doIdleGCWork(cap, true /* all of it */);
    }

    traceSparkCounters(cap);

    switch (SEQ_CST_LOAD(&recent_activity)) {
    case ACTIVITY_INACTIVE:
        if (force_major) {
            // We are doing a GC because the system has been idle for a
            // timeslice and we need to check for deadlock.  Record the
            // fact that we've done a GC and turn off the timer signal;
            // it will get re-enabled if we run any threads after the GC.
            SEQ_CST_STORE(&recent_activity, ACTIVITY_DONE_GC);
#if !defined(PROFILING)
            stopTimer();
#endif
            break;
        }
        // fall through...

    case ACTIVITY_MAYBE_NO:
        // the GC might have taken long enough for the timer to set
        // recent_activity = ACTIVITY_MAYBE_NO or ACTIVITY_INACTIVE,
        // but we aren't necessarily deadlocked:
        SEQ_CST_STORE(&recent_activity, ACTIVITY_YES);
        break;

    case ACTIVITY_DONE_GC:
        // If we are actually active, the scheduler will reset the
        // recent_activity flag and re-enable the timer.
        break;
    }

#if defined(THREADED_RTS)
    // Stable point where we can do a global check on our spark counters
    ASSERT(checkSparkCountInvariant());
#endif

    // The heap census itself is done during GarbageCollect().
    if (heap_census) {
        performHeapProfile = false;
    }

#if defined(THREADED_RTS)

    // If n_capabilities has changed during GC, we're in trouble.
    ASSERT(n_capabilities == old_n_capabilities);

    if (gc_type == SYNC_GC_PAR)
    {
        for (i = 0; i < n_capabilities; i++) {
            if (i != cap->no) {
                if (idle_cap[i]) {
                    ASSERT(capabilities[i]->running_task == task);
                    task->cap = capabilities[i];
                    releaseCapability(capabilities[i]);
                } else {
                    ASSERT(capabilities[i]->running_task != task);
                }
            }
        }
        task->cap = cap;

        // releaseGCThreads() happens *after* we have released idle
        // capabilities.  Otherwise what can happen is one of the released
        // threads starts a new GC, and finds that it can't acquire some of
        // the disabled capabilities, because the previous GC still holds
        // them, so those disabled capabilities will not be idle during the
        // next GC round.  However, if we release the capabilities first,
        // then they will be free (because they're disabled) when the next
        // GC cycle happens.
        releaseGCThreads(cap, idle_cap);
    }
#endif
    if (heap_overflow && RELAXED_LOAD(&sched_state) == SCHED_RUNNING) {
        // GC set the heap_overflow flag.  We should throw an exception if we
        // can, or shut down otherwise.

        // Get the thread to which Ctrl-C is thrown
        StgTSO *main_thread = getTopHandlerThread();
        if (main_thread == NULL) {
            // GC set the heap_overflow flag, and there is no main thread to
            // throw an exception to, so we should proceed with an orderly
            // shutdown now.  Ultimately we want the main thread to return to
            // its caller with HeapExhausted, at which point the caller should
            // call hs_exit().  The first step is to delete all the threads.
            RELAXED_STORE(&sched_state, SCHED_INTERRUPTING);
            goto delete_threads_and_gc;
        }

        heap_overflow = false;
        const uint64_t allocation_count = getAllocations();
        if (RtsFlags.GcFlags.heapLimitGrace <
              allocation_count - allocated_bytes_at_heapoverflow ||
              allocated_bytes_at_heapoverflow == 0) {
            allocated_bytes_at_heapoverflow = allocation_count;
            // We used to simply exit, but throwing an exception gives the
            // program a chance to clean up.  It also lets the exception be
            // caught.

            // FIXME this is not a good way to tell a program to release
            // resources.  It is neither reliable (the RTS crashes if it fails
            // to allocate memory from the OS) nor very usable (it is always
            // thrown to the main thread, which might not be able to do anything
            // useful with it).  We really should have a more general way to
            // release resources in low-memory conditions.  Nevertheless, this
            // is still a big improvement over just exiting.

            // FIXME again: perhaps we should throw a synchronous exception
            // instead an asynchronous one, or have a way for the program to
            // register a handler to be called when heap overflow happens.
            throwToSelf(cap, main_thread, heapOverflow_closure);
        }
    }

#if defined(THREADED_RTS)
    stgFree(idle_cap);

    if (gc_type == SYNC_GC_SEQ) {
        // release our stash of capabilities.
        releaseAllCapabilities(n_capabilities, cap, task);
    }
#endif

    return;
}

/* ---------------------------------------------------------------------------
 * Singleton fork(). Do not copy any running threads.
 * ------------------------------------------------------------------------- */

pid_t
forkProcess(HsStablePtr *entry
#if !defined(FORKPROCESS_PRIMOP_SUPPORTED)
            STG_UNUSED
#endif
           )
{
#if defined(FORKPROCESS_PRIMOP_SUPPORTED)
    pid_t pid;
    StgTSO* t,*next;
    Capability *cap;
    uint32_t g;
    Task *task = NULL;
    uint32_t i;

    debugTrace(DEBUG_sched, "forking!");

    task = newBoundTask();

    cap = NULL;
    waitForCapability(&cap, task);

#if defined(THREADED_RTS)
    stopAllCapabilities(&cap, task);
#endif

    // no funny business: hold locks while we fork, otherwise if some
    // other thread is holding a lock when the fork happens, the data
    // structure protected by the lock will forever be in an
    // inconsistent state in the child.  See also #1391.
    ACQUIRE_LOCK(&sched_mutex);
    ACQUIRE_LOCK(&sm_mutex);
    ACQUIRE_LOCK(&stable_ptr_mutex);
    ACQUIRE_LOCK(&stable_name_mutex);

    for (i=0; i < n_capabilities; i++) {
        ACQUIRE_LOCK(&capabilities[i]->lock);
    }

    // Take task lock after capability lock to avoid order inversion (#17275).
    ACQUIRE_LOCK(&task->lock);

#if defined(THREADED_RTS)
    ACQUIRE_LOCK(&all_tasks_mutex);
#endif

    stopTimer(); // See #4074

#if defined(TRACING)
    flushAllCapsEventsBufs(); // so that child won't inherit dirty file buffers
#endif

    pid = fork();

    if (pid) { // parent

        startTimer(); // #4074

        RELEASE_LOCK(&sched_mutex);
        RELEASE_LOCK(&sm_mutex);
        RELEASE_LOCK(&stable_ptr_mutex);
        RELEASE_LOCK(&stable_name_mutex);
        RELEASE_LOCK(&task->lock);

#if defined(THREADED_RTS)
        /* N.B. releaseCapability_ below may need to take all_tasks_mutex */
        RELEASE_LOCK(&all_tasks_mutex);
#endif

        for (i=0; i < n_capabilities; i++) {
            releaseCapability_(capabilities[i],false);
            RELEASE_LOCK(&capabilities[i]->lock);
        }

        exitMyTask();

        // just return the pid
        return pid;

    } else { // child

        // Current process times reset in the child process, so we should reset
        // the stats too. See #16102.
        resetChildProcessStats();

#if defined(THREADED_RTS)
        initMutex(&sched_mutex);
        initMutex(&sm_mutex);
        initMutex(&stable_ptr_mutex);
        initMutex(&stable_name_mutex);
        initMutex(&task->lock);

        for (i=0; i < n_capabilities; i++) {
            initMutex(&capabilities[i]->lock);
        }

        initMutex(&all_tasks_mutex);
#endif

#if defined(TRACING)
        resetTracing();
#endif

        // Now, all OS threads except the thread that forked are
        // stopped.  We need to stop all Haskell threads, including
        // those involved in foreign calls.  Also we need to delete
        // all Tasks, because they correspond to OS threads that are
        // now gone.

        for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
          for (t = generations[g].threads; t != END_TSO_QUEUE; t = next) {
                next = t->global_link;
                // don't allow threads to catch the ThreadKilled
                // exception, but we do want to raiseAsync() because these
                // threads may be evaluating thunks that we need later.
                deleteThread_(t);

                // stop the GC from updating the InCall to point to
                // the TSO.  This is only necessary because the
                // OSThread bound to the TSO has been killed, and
                // won't get a chance to exit in the usual way (see
                // also scheduleHandleThreadFinished).
                t->bound = NULL;
          }
        }

        discardTasksExcept(task);

        for (i=0; i < n_capabilities; i++) {
            cap = capabilities[i];

            // Empty the run queue.  It seems tempting to let all the
            // killed threads stay on the run queue as zombies to be
            // cleaned up later, but some of them may correspond to
            // bound threads for which the corresponding Task does not
            // exist.
            truncateRunQueue(cap);
            cap->n_run_queue = 0;

            // Any suspended C-calling Tasks are no more, their OS threads
            // don't exist now:
            cap->suspended_ccalls = NULL;
            cap->n_suspended_ccalls = 0;

#if defined(THREADED_RTS)
            // Wipe our spare workers list, they no longer exist.  New
            // workers will be created if necessary.
            cap->spare_workers = NULL;
            cap->n_spare_workers = 0;
            cap->returning_tasks_hd = NULL;
            cap->returning_tasks_tl = NULL;
            cap->n_returning_tasks = 0;
#endif

            // Release all caps except 0, we'll use that for starting
            // the IO manager and running the client action below.
            if (cap->no != 0) {
                task->cap = cap;
                releaseCapability(cap);
            }
        }
        cap = capabilities[0];
        task->cap = cap;

        // Empty the threads lists.  Otherwise, the garbage
        // collector may attempt to resurrect some of these threads.
        for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
            generations[g].threads = END_TSO_QUEUE;
        }

        // On Unix, all timers are reset in the child, so we need to start
        // the timer again.
        initTimer();

        // TODO: need to trace various other things in the child
        // like startup event, capabilities, process info etc
        traceTaskCreate(task, cap);

        initIOManagerAfterFork(&cap);

        // start timer after the IOManager is initialized
        // (the idle GC may wake up the IOManager)
        startTimer();

        // Install toplevel exception handlers, so interruption
        // signal will be sent to the main thread.
        // See #12903
        rts_evalStableIOMain(&cap, entry, NULL);  // run the action
        rts_checkSchedStatus("forkProcess",cap);

        rts_unlock(cap);
        shutdownHaskellAndExit(EXIT_SUCCESS, 0 /* !fastExit */);
    }
#else /* !FORKPROCESS_PRIMOP_SUPPORTED */
    barf("forkProcess#: primop not supported on this platform, sorry!\n");
#endif
}

/* ---------------------------------------------------------------------------
 * Changing the number of Capabilities
 *
 * Changing the number of Capabilities is very tricky!  We can only do
 * it with the system fully stopped, so we do a full sync with
 * requestSync(SYNC_OTHER) and grab all the capabilities.
 *
 * Then we resize the appropriate data structures, and update all
 * references to the old data structures which have now moved.
 * Finally we release the Capabilities we are holding, and start
 * worker Tasks on the new Capabilities we created.
 *
 * ------------------------------------------------------------------------- */

void
setNumCapabilities (uint32_t new_n_capabilities USED_IF_THREADS)
{
#if !defined(THREADED_RTS)
    if (new_n_capabilities != 1) {
        errorBelch("setNumCapabilities: not supported in the non-threaded RTS");
    }
    return;
#elif defined(NOSMP)
    if (new_n_capabilities != 1) {
        errorBelch("setNumCapabilities: not supported on this platform");
    }
    return;
#else
    Task *task;
    Capability *cap;
    uint32_t n;
    Capability *old_capabilities = NULL;
    uint32_t old_n_capabilities = n_capabilities;

    if (new_n_capabilities == enabled_capabilities) {
        return;
    } else if (new_n_capabilities <= 0) {
        errorBelch("setNumCapabilities: Capability count must be positive");
        return;
    }


    debugTrace(DEBUG_sched, "changing the number of Capabilities from %d to %d",
               enabled_capabilities, new_n_capabilities);

    cap = rts_lock();
    task = cap->running_task;


    // N.B. We must stop the interval timer while we are changing the
    // capabilities array lest handle_tick may try to context switch
    // an old capability. See #17289.
    stopTimer();

    stopAllCapabilities(&cap, task);

    if (new_n_capabilities < enabled_capabilities)
    {
        // Reducing the number of capabilities: we do not actually
        // remove the extra capabilities, we just mark them as
        // "disabled". This has the following effects:
        //
        //   - threads on a disabled capability are migrated away by the
        //     scheduler loop
        //
        //   - disabled capabilities do not participate in GC
        //     (see scheduleDoGC())
        //
        //   - No spark threads are created on this capability
        //     (see scheduleActivateSpark())
        //
        //   - We do not attempt to migrate threads *to* a disabled
        //     capability (see schedulePushWork()).
        //
        // but in other respects, a disabled capability remains
        // alive.  Threads may be woken up on a disabled capability,
        // but they will be immediately migrated away.
        //
        // This approach is much easier than trying to actually remove
        // the capability; we don't have to worry about GC data
        // structures, the nursery, etc.
        //
        for (n = new_n_capabilities; n < enabled_capabilities; n++) {
            capabilities[n]->disabled = true;
            traceCapDisable(capabilities[n]);
        }
        enabled_capabilities = new_n_capabilities;
    }
    else
    {
        // Increasing the number of enabled capabilities.
        //
        // enable any disabled capabilities, up to the required number
        for (n = enabled_capabilities;
             n < new_n_capabilities && n < n_capabilities; n++) {
            capabilities[n]->disabled = false;
            traceCapEnable(capabilities[n]);
        }
        enabled_capabilities = n;

        if (new_n_capabilities > n_capabilities) {
#if defined(TRACING)
            // Allocate eventlog buffers for the new capabilities.  Note this
            // must be done before calling moreCapabilities(), because that
            // will emit events about creating the new capabilities and adding
            // them to existing capsets.
            tracingAddCapapilities(n_capabilities, new_n_capabilities);
#endif

            // Resize the capabilities array
            // NB. after this, capabilities points somewhere new.  Any pointers
            // of type (Capability *) are now invalid.
            moreCapabilities(n_capabilities, new_n_capabilities);

            // Resize and update storage manager data structures
            storageAddCapabilities(n_capabilities, new_n_capabilities);
        }
    }

    // update n_capabilities before things start running
    if (new_n_capabilities > n_capabilities) {
        n_capabilities = enabled_capabilities = new_n_capabilities;
    }

    // We're done: release the original Capabilities
    releaseAllCapabilities(old_n_capabilities, cap,task);

    // We can't free the old array until now, because we access it
    // while updating pointers in updateCapabilityRefs().
    if (old_capabilities) {
        stgFree(old_capabilities);
    }

    // Notify IO manager that the number of capabilities has changed.
    rts_evalIO(&cap, ioManagerCapabilitiesChanged_closure, NULL);

    startTimer();

    rts_unlock(cap);

#endif // THREADED_RTS
}



/* ---------------------------------------------------------------------------
 * Delete all the threads in the system
 * ------------------------------------------------------------------------- */

static void
deleteAllThreads ()
{
    // NOTE: only safe to call if we own all capabilities.

    StgTSO* t, *next;
    uint32_t g;

    debugTrace(DEBUG_sched,"deleting all threads");
    for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
        for (t = generations[g].threads; t != END_TSO_QUEUE; t = next) {
                next = t->global_link;
                deleteThread(t);
        }
    }

    // The run queue now contains a bunch of ThreadKilled threads.  We
    // must not throw these away: the main thread(s) will be in there
    // somewhere, and the main scheduler loop has to deal with it.
    // Also, the run queue is the only thing keeping these threads from
    // being GC'd, and we don't want the "main thread has been GC'd" panic.

#if !defined(THREADED_RTS)
    ASSERT(blocked_queue_hd == END_TSO_QUEUE);
    ASSERT(sleeping_queue == END_TSO_QUEUE);
#endif
}

/* -----------------------------------------------------------------------------
   Managing the suspended_ccalls list.
   Locks required: sched_mutex
   -------------------------------------------------------------------------- */

STATIC_INLINE void
suspendTask (Capability *cap, Task *task)
{
    InCall *incall;

    incall = task->incall;
    ASSERT(incall->next == NULL && incall->prev == NULL);
    incall->next = cap->suspended_ccalls;
    incall->prev = NULL;
    if (cap->suspended_ccalls) {
        cap->suspended_ccalls->prev = incall;
    }
    cap->suspended_ccalls = incall;
    cap->n_suspended_ccalls++;
}

STATIC_INLINE void
recoverSuspendedTask (Capability *cap, Task *task)
{
    InCall *incall;

    incall = task->incall;
    if (incall->prev) {
        incall->prev->next = incall->next;
    } else {
        ASSERT(cap->suspended_ccalls == incall);
        cap->suspended_ccalls = incall->next;
    }
    if (incall->next) {
        incall->next->prev = incall->prev;
    }
    incall->next = incall->prev = NULL;
    cap->n_suspended_ccalls--;
}

/* ---------------------------------------------------------------------------
 * Suspending & resuming Haskell threads.
 *
 * When making a "safe" call to C (aka _ccall_GC), the task gives back
 * its capability before calling the C function.  This allows another
 * task to pick up the capability and carry on running Haskell
 * threads.  It also means that if the C call blocks, it won't lock
 * the whole system.
 *
 * The Haskell thread making the C call is put to sleep for the
 * duration of the call, on the suspended_ccalling_threads queue.  We
 * give out a token to the task, which it can use to resume the thread
 * on return from the C function.
 *
 * If this is an interruptible C call, this means that the FFI call may be
 * unceremoniously terminated and should be scheduled on an
 * unbound worker thread.
 * ------------------------------------------------------------------------- */

void *
suspendThread (StgRegTable *reg, bool interruptible)
{
  Capability *cap;
  int saved_errno;
  StgTSO *tso;
  Task *task;
#if defined(mingw32_HOST_OS)
  StgWord32 saved_winerror;
#endif

  saved_errno = errno;
#if defined(mingw32_HOST_OS)
  saved_winerror = GetLastError();
#endif

  /* assume that *reg is a pointer to the StgRegTable part of a Capability.
   */
  cap = regTableToCapability(reg);

  task = cap->running_task;
  tso = cap->r.rCurrentTSO;

  traceEventStopThread(cap, tso, THREAD_SUSPENDED_FOREIGN_CALL, 0);

  // XXX this might not be necessary --SDM
  tso->what_next = ThreadRunGHC;

  threadPaused(cap,tso);

  if (interruptible) {
    tso->why_blocked = BlockedOnCCall_Interruptible;
  } else {
    tso->why_blocked = BlockedOnCCall;
  }

  // Hand back capability
  task->incall->suspended_tso = tso;
  task->incall->suspended_cap = cap;

  // Otherwise allocate() will write to invalid memory.
  cap->r.rCurrentTSO = NULL;

  ACQUIRE_LOCK(&cap->lock);

  suspendTask(cap,task);
  cap->in_haskell = false;
  releaseCapability_(cap,false);

  RELEASE_LOCK(&cap->lock);

  errno = saved_errno;
#if defined(mingw32_HOST_OS)
  SetLastError(saved_winerror);
#endif
  return task;
}

StgRegTable *
resumeThread (void *task_)
{
    StgTSO *tso;
    InCall *incall;
    Capability *cap;
    Task *task = task_;
    int saved_errno;
#if defined(mingw32_HOST_OS)
    StgWord32 saved_winerror;
#endif

    saved_errno = errno;
#if defined(mingw32_HOST_OS)
    saved_winerror = GetLastError();
#endif

    incall = task->incall;
    cap = incall->suspended_cap;
    task->cap = cap;

    // Wait for permission to re-enter the RTS with the result.
    waitForCapability(&cap,task);
    // we might be on a different capability now... but if so, our
    // entry on the suspended_ccalls list will also have been
    // migrated.

    // Remove the thread from the suspended list
    recoverSuspendedTask(cap,task);

    tso = incall->suspended_tso;
    incall->suspended_tso = NULL;
    incall->suspended_cap = NULL;
    // we will modify tso->_link
    IF_NONMOVING_WRITE_BARRIER_ENABLED {
        updateRemembSetPushClosure(cap, (StgClosure *)tso->_link);
    }
    tso->_link = END_TSO_QUEUE;

    traceEventRunThread(cap, tso);

    /* Reset blocking status */
    tso->why_blocked  = NotBlocked;

    if ((tso->flags & TSO_BLOCKEX) == 0) {
        // avoid locking the TSO if we don't have to
        if (tso->blocked_exceptions != END_BLOCKED_EXCEPTIONS_QUEUE) {
            maybePerformBlockedException(cap,tso);
        }
    }

    cap->r.rCurrentTSO = tso;
    cap->in_haskell = true;
    errno = saved_errno;
#if defined(mingw32_HOST_OS)
    SetLastError(saved_winerror);
#endif

    /* We might have GC'd, mark the TSO dirty again */
    dirty_TSO(cap,tso);
    dirty_STACK(cap,tso->stackobj);

    IF_DEBUG(sanity, checkTSO(tso));

    return &cap->r;
}

/* ---------------------------------------------------------------------------
 * scheduleThread()
 *
 * scheduleThread puts a thread on the end  of the runnable queue.
 * This will usually be done immediately after a thread is created.
 * The caller of scheduleThread must create the thread using e.g.
 * createThread and push an appropriate closure
 * on this thread's stack before the scheduler is invoked.
 * ------------------------------------------------------------------------ */

void
scheduleThread(Capability *cap, StgTSO *tso)
{
    // The thread goes at the *end* of the run-queue, to avoid possible
    // starvation of any threads already on the queue.
    appendToRunQueue(cap,tso);
}

void
scheduleThreadNow(Capability *cap, StgTSO *tso)
{
    // The thread goes at the *beginning* of the run-queue,
    // in order to execute it as soon as possible.
    pushOnRunQueue(cap,tso);
}

void
scheduleThreadOn(Capability *cap, StgWord cpu USED_IF_THREADS, StgTSO *tso)
{
    tso->flags |= TSO_LOCKED; // we requested explicit affinity; don't
                              // move this thread from now on.
#if defined(THREADED_RTS)
    cpu %= enabled_capabilities;
    if (cpu == cap->no) {
        appendToRunQueue(cap,tso);
    } else {
        migrateThread(cap, tso, capabilities[cpu]);
    }
#else
    appendToRunQueue(cap,tso);
#endif
}

// See rts/include/rts/Threads.h
void
scheduleWaitThread (StgTSO* tso, /*[out]*/HaskellObj* ret, Capability **pcap)
{
    Task *task;
    DEBUG_ONLY( StgThreadID id );
    Capability *cap;

    cap = *pcap;

    // We already created/initialised the Task
    task = cap->running_task;

    // This TSO is now a bound thread; make the Task and TSO
    // point to each other.
    tso->bound = task->incall;
    tso->cap = cap;

    task->incall->tso = tso;
    task->incall->ret = ret;
    task->incall->rstat = NoStatus;

    appendToRunQueue(cap,tso);

    DEBUG_ONLY( id = tso->id );
    debugTrace(DEBUG_sched, "new bound thread (%" FMT_StgThreadID ")", id);

    // As the TSO is bound and on the run queue, schedule() will run the TSO.
    cap = schedule(cap,task);

    ASSERT(task->incall->rstat != NoStatus);
    ASSERT_FULL_CAPABILITY_INVARIANTS(cap,task);

    debugTrace(DEBUG_sched, "bound thread (%" FMT_StgThreadID ") finished", id);
    *pcap = cap;
}

/* ----------------------------------------------------------------------------
 * Starting Tasks
 * ------------------------------------------------------------------------- */

#if defined(THREADED_RTS)
void scheduleWorker (Capability *cap, Task *task)
{
    ASSERT_FULL_CAPABILITY_INVARIANTS(cap, task);
    cap = schedule(cap,task);
    ASSERT_FULL_CAPABILITY_INVARIANTS(cap, task);

    // On exit from schedule(), we have a Capability, but possibly not
    // the same one we started with.

    // During shutdown, the requirement is that after all the
    // Capabilities are shut down, all workers that are shutting down
    // have finished workerTaskStop().  This is why we hold on to
    // cap->lock until we've finished workerTaskStop() below.
    //
    // There may be workers still involved in foreign calls; those
    // will just block in waitForCapability() because the
    // Capability has been shut down.
    //
    ACQUIRE_LOCK(&cap->lock);
    releaseCapability_(cap,false);
    workerTaskStop(task);
    RELEASE_LOCK(&cap->lock);
}
#endif

/* ---------------------------------------------------------------------------
 * Start new worker tasks on Capabilities from--to
 * -------------------------------------------------------------------------- */

static void
startWorkerTasks (uint32_t from USED_IF_THREADS, uint32_t to USED_IF_THREADS)
{
#if defined(THREADED_RTS)
    uint32_t i;
    Capability *cap;

    for (i = from; i < to; i++) {
        cap = capabilities[i];
        ACQUIRE_LOCK(&cap->lock);
        startWorkerTask(cap);
        RELEASE_LOCK(&cap->lock);
    }
#endif
}

/* ---------------------------------------------------------------------------
 * initScheduler()
 *
 * Initialise the scheduler.  This resets all the queues - if the
 * queues contained any threads, they'll be garbage collected at the
 * next pass.
 *
 * ------------------------------------------------------------------------ */

void
initScheduler(void)
{
#if !defined(THREADED_RTS)
  blocked_queue_hd  = END_TSO_QUEUE;
  blocked_queue_tl  = END_TSO_QUEUE;
  sleeping_queue    = END_TSO_QUEUE;
#endif

  sched_state    = SCHED_RUNNING;
  SEQ_CST_STORE(&recent_activity, ACTIVITY_YES);


  /* Initialise the mutex and condition variables used by
   * the scheduler. */
#if defined(THREADED_RTS)
  initMutex(&sched_mutex);
  initMutex(&sync_finished_mutex);
  initCondition(&sync_finished_cond);
#endif

  ACQUIRE_LOCK(&sched_mutex);

  allocated_bytes_at_heapoverflow = 0;

  /* A capability holds the state a native thread needs in
   * order to execute STG code. At least one capability is
   * floating around (only THREADED_RTS builds have more than one).
   */
  initCapabilities();

  initTaskManager();

  /*
   * Eagerly start one worker to run each Capability, except for
   * Capability 0.  The idea is that we're probably going to start a
   * bound thread on Capability 0 pretty soon, so we don't want a
   * worker task hogging it.
   */
  startWorkerTasks(1, n_capabilities);

  RELEASE_LOCK(&sched_mutex);

}

void
exitScheduler (bool wait_foreign USED_IF_THREADS)
               /* see Capability.c, shutdownCapability() */
{
    Task *task = newBoundTask();

    // If we haven't killed all the threads yet, do it now.
    if (RELAXED_LOAD(&sched_state) < SCHED_SHUTTING_DOWN) {
        RELAXED_STORE(&sched_state, SCHED_INTERRUPTING);
        nonmovingStop();
        Capability *cap = task->cap;
        waitForCapability(&cap,task);
        scheduleDoGC(&cap,task,true,false,false);
        ASSERT(task->incall->tso == NULL);
        releaseCapability(cap);
    }
    ASSERT(sched_state == SCHED_SHUTTING_DOWN);

    shutdownCapabilities(task, wait_foreign);

    // debugBelch("n_failed_trygrab_idles = %d, n_idle_caps = %d\n",
    //            n_failed_trygrab_idles, n_idle_caps);

    exitMyTask();
}

void
freeScheduler( void )
{
    uint32_t still_running;

    ACQUIRE_LOCK(&sched_mutex);
    still_running = freeTaskManager();
    // We can only free the Capabilities if there are no Tasks still
    // running.  We might have a Task about to return from a foreign
    // call into waitForCapability(), for example (actually,
    // this should be the *only* thing that a still-running Task can
    // do at this point, and it will block waiting for the
    // Capability).
    if (still_running == 0) {
        freeCapabilities();
    }
    RELEASE_LOCK(&sched_mutex);
#if defined(THREADED_RTS)
    closeMutex(&sched_mutex);
#endif
}

void markScheduler (evac_fn evac USED_IF_NOT_THREADS,
                    void *user USED_IF_NOT_THREADS)
{
#if !defined(THREADED_RTS)
    evac(user, (StgClosure **)(void *)&blocked_queue_hd);
    evac(user, (StgClosure **)(void *)&blocked_queue_tl);
    evac(user, (StgClosure **)(void *)&sleeping_queue);
#endif
}

/* -----------------------------------------------------------------------------
   performGC

   This is the interface to the garbage collector from Haskell land.
   We provide this so that external C code can allocate and garbage
   collect when called from Haskell via _ccall_GC.
   -------------------------------------------------------------------------- */

static void
performGC_(bool force_major)
{
    Task *task;
    Capability *cap = NULL;

    // We must grab a new Task here, because the existing Task may be
    // associated with a particular Capability, and chained onto the
    // suspended_ccalls queue.
    task = newBoundTask();

    // TODO: do we need to traceTask*() here?

    waitForCapability(&cap,task);
    scheduleDoGC(&cap,task,force_major,false,false);
    releaseCapability(cap);
    exitMyTask();
}

void
performGC(void)
{
    performGC_(false);
}

void
performMajorGC(void)
{
    performGC_(true);
}

/* ---------------------------------------------------------------------------
   Interrupt execution.
   Might be called inside a signal handler so it mustn't do anything fancy.
   ------------------------------------------------------------------------ */

void
interruptStgRts(void)
{
    ASSERT(sched_state != SCHED_SHUTTING_DOWN);
    sched_state = SCHED_INTERRUPTING;
    interruptAllCapabilities();
#if defined(THREADED_RTS)
    wakeUpRts();
#endif
}

/* -----------------------------------------------------------------------------
   Wake up the RTS

   This function causes at least one OS thread to wake up and run the
   scheduler loop.  It is invoked when the RTS might be deadlocked, or
   an external event has arrived that may need servicing (eg. a
   keyboard interrupt).

   In the single-threaded RTS we don't do anything here; we only have
   one thread anyway, and the event that caused us to want to wake up
   will have interrupted any blocking system call in progress anyway.
   -------------------------------------------------------------------------- */

#if defined(THREADED_RTS)
void wakeUpRts(void)
{
    // This forces the IO Manager thread to wakeup, which will
    // in turn ensure that some OS thread wakes up and runs the
    // scheduler loop, which will cause a GC and deadlock check.
    wakeupIOManager();
}
#endif

/* -----------------------------------------------------------------------------
   Deleting threads

   This is used for interruption (^C) and forking, and corresponds to
   raising an exception but without letting the thread catch the
   exception.
   -------------------------------------------------------------------------- */

static void
deleteThread (StgTSO *tso)
{
    // NOTE: must only be called on a TSO that we have exclusive
    // access to, because we will call throwToSingleThreaded() below.
    // The TSO must be on the run queue of the Capability we own, or
    // we must own all Capabilities.

    if (tso->why_blocked != BlockedOnCCall &&
        tso->why_blocked != BlockedOnCCall_Interruptible) {
        throwToSingleThreaded(tso->cap,tso,NULL);
    }
}

#if defined(FORKPROCESS_PRIMOP_SUPPORTED)
static void
deleteThread_(StgTSO *tso)
{ // for forkProcess only:
  // like deleteThread(), but we delete threads in foreign calls, too.

    if (tso->why_blocked == BlockedOnCCall ||
        tso->why_blocked == BlockedOnCCall_Interruptible) {
        tso->what_next = ThreadKilled;
        appendToRunQueue(tso->cap, tso);
    } else {
        deleteThread(tso);
    }
}
#endif

/* -----------------------------------------------------------------------------
   raiseExceptionHelper

   This function is called by the raise# primitive, just so that we can
   move some of the tricky bits of raising an exception from C-- into
   C.  Who knows, it might be a useful re-useable thing here too.
   -------------------------------------------------------------------------- */

StgWord
raiseExceptionHelper (StgRegTable *reg, StgTSO *tso, StgClosure *exception)
{
    Capability *cap = regTableToCapability(reg);
    StgThunk *raise_closure = NULL;
    StgPtr p, next;
    const StgRetInfoTable *info;
    //
    // This closure represents the expression 'raise# E' where E
    // is the exception raise.  It is used to overwrite all the
    // thunks which are currently under evaluation.
    //

    // OLD COMMENT (we don't have MIN_UPD_SIZE now):
    // LDV profiling: stg_raise_info has THUNK as its closure
    // type. Since a THUNK takes at least MIN_UPD_SIZE words in its
    // payload, MIN_UPD_SIZE is more appropriate than 1.  It seems that
    // 1 does not cause any problem unless profiling is performed.
    // However, when LDV profiling goes on, we need to linearly scan
    // small object pool, where raise_closure is stored, so we should
    // use MIN_UPD_SIZE.
    //
    // raise_closure = (StgClosure *)RET_STGCALL1(P_,allocate,
    //                                 sizeofW(StgClosure)+1);
    //

    //
    // Walk up the stack, looking for the catch frame.  On the way,
    // we update any closures pointed to from update frames with the
    // raise closure that we just built.
    //
    p = tso->stackobj->sp;
    while(1) {
        info = get_ret_itbl((StgClosure *)p);
        next = p + stack_frame_sizeW((StgClosure *)p);
        switch (info->i.type) {

        case UPDATE_FRAME:
            // Only create raise_closure if we need to.
            if (raise_closure == NULL) {
                raise_closure =
                    (StgThunk *)allocate(cap,sizeofW(StgThunk)+1);
                SET_HDR(raise_closure, &stg_raise_info, cap->r.rCCCS);
                raise_closure->payload[0] = exception;
            }
            updateThunk(cap, tso, ((StgUpdateFrame *)p)->updatee,
                        (StgClosure *)raise_closure);
            p = next;
            continue;

        case ATOMICALLY_FRAME:
            debugTrace(DEBUG_stm, "found ATOMICALLY_FRAME at %p", p);
            tso->stackobj->sp = p;
            return ATOMICALLY_FRAME;

        case CATCH_FRAME:
            tso->stackobj->sp = p;
            return CATCH_FRAME;

        case CATCH_STM_FRAME:
            debugTrace(DEBUG_stm, "found CATCH_STM_FRAME at %p", p);
            tso->stackobj->sp = p;
            return CATCH_STM_FRAME;

        case UNDERFLOW_FRAME:
            tso->stackobj->sp = p;
            threadStackUnderflow(cap,tso);
            p = tso->stackobj->sp;
            continue;

        case STOP_FRAME:
            tso->stackobj->sp = p;
            return STOP_FRAME;

        case CATCH_RETRY_FRAME: {
            StgTRecHeader *trec = tso -> trec;
            StgTRecHeader *outer = trec -> enclosing_trec;
            debugTrace(DEBUG_stm,
                       "found CATCH_RETRY_FRAME at %p during raise", p);
            debugTrace(DEBUG_stm, "trec=%p outer=%p", trec, outer);
            stmAbortTransaction(cap, trec);
            stmFreeAbortedTRec(cap, trec);
            tso -> trec = outer;
            p = next;
            continue;
        }

        default:
            p = next;
            continue;
        }
    }
}


/* -----------------------------------------------------------------------------
   findRetryFrameHelper

   This function is called by the retry# primitive.  It traverses the stack
   leaving tso->sp referring to the frame which should handle the retry.

   This should either be a CATCH_RETRY_FRAME (if the retry# is within an orElse#)
   or should be a ATOMICALLY_FRAME (if the retry# reaches the top level).

   We skip CATCH_STM_FRAMEs (aborting and rolling back the nested tx that they
   create) because retries are not considered to be exceptions, despite the
   similar implementation.

   We should not expect to see CATCH_FRAME or STOP_FRAME because those should
   not be created within memory transactions.
   -------------------------------------------------------------------------- */

StgWord
findRetryFrameHelper (Capability *cap, StgTSO *tso)
{
  const StgRetInfoTable *info;
  StgPtr    p, next;

  p = tso->stackobj->sp;
  while (1) {
    info = get_ret_itbl((const StgClosure *)p);
    next = p + stack_frame_sizeW((StgClosure *)p);
    switch (info->i.type) {

    case ATOMICALLY_FRAME:
        debugTrace(DEBUG_stm,
                   "found ATOMICALLY_FRAME at %p during retry", p);
        tso->stackobj->sp = p;
        return ATOMICALLY_FRAME;

    case CATCH_RETRY_FRAME:
        debugTrace(DEBUG_stm,
                   "found CATCH_RETRY_FRAME at %p during retry", p);
        tso->stackobj->sp = p;
        return CATCH_RETRY_FRAME;

    case CATCH_STM_FRAME: {
        StgTRecHeader *trec = tso -> trec;
        StgTRecHeader *outer = trec -> enclosing_trec;
        debugTrace(DEBUG_stm,
                   "found CATCH_STM_FRAME at %p during retry", p);
        debugTrace(DEBUG_stm, "trec=%p outer=%p", trec, outer);
        stmAbortTransaction(cap, trec);
        stmFreeAbortedTRec(cap, trec);
        tso -> trec = outer;
        p = next;
        continue;
    }

    case UNDERFLOW_FRAME:
        tso->stackobj->sp = p;
        threadStackUnderflow(cap,tso);
        p = tso->stackobj->sp;
        continue;

    default:
      ASSERT(info->i.type != CATCH_FRAME);
      ASSERT(info->i.type != STOP_FRAME);
      p = next;
      continue;
    }
  }
}

/* -----------------------------------------------------------------------------
   findAtomicallyFrameHelper

   This function is called by stg_abort via catch_retry_frame primitive.  It is
   like findRetryFrameHelper but it will only stop at ATOMICALLY_FRAME.
   -------------------------------------------------------------------------- */

StgWord
findAtomicallyFrameHelper (Capability *cap, StgTSO *tso)
{
  const StgRetInfoTable *info;
  StgPtr    p, next;

  p = tso->stackobj->sp;
  while (1) {
    info = get_ret_itbl((const StgClosure *)p);
    next = p + stack_frame_sizeW((StgClosure *)p);
    switch (info->i.type) {

    case ATOMICALLY_FRAME:
        debugTrace(DEBUG_stm,
                   "found ATOMICALLY_FRAME at %p while aborting after orElse", p);
        tso->stackobj->sp = p;
        return ATOMICALLY_FRAME;

    case CATCH_RETRY_FRAME: {
        StgTRecHeader *trec = tso -> trec;
        StgTRecHeader *outer = trec -> enclosing_trec;
        debugTrace(DEBUG_stm,
                   "found CATCH_RETRY_FRAME at %p while aborting after orElse", p);
        debugTrace(DEBUG_stm, "trec=%p outer=%p", trec, outer);
        stmAbortTransaction(cap, trec);
        stmFreeAbortedTRec(cap, trec);
        tso -> trec = outer;
        p = next;
        continue;
    }

    case CATCH_STM_FRAME: {
        StgTRecHeader *trec = tso -> trec;
        StgTRecHeader *outer = trec -> enclosing_trec;
        debugTrace(DEBUG_stm,
                   "found CATCH_STM_FRAME at %p while aborting after orElse", p);
        debugTrace(DEBUG_stm, "trec=%p outer=%p", trec, outer);
        stmAbortTransaction(cap, trec);
        stmFreeAbortedTRec(cap, trec);
        tso -> trec = outer;
        p = next;
        continue;
    }

    case UNDERFLOW_FRAME:
        tso->stackobj->sp = p;
        threadStackUnderflow(cap,tso);
        p = tso->stackobj->sp;
        continue;

    default:
      ASSERT(info->i.type != CATCH_FRAME);
      ASSERT(info->i.type != STOP_FRAME);
      p = next;
      continue;
    }
  }
}

/* -----------------------------------------------------------------------------
   resurrectThreads is called after garbage collection on the list of
   threads found to be garbage.  Each of these threads will be woken
   up and sent a signal: BlockedOnDeadMVar if the thread was blocked
   on an MVar, or NonTermination if the thread was blocked on a Black
   Hole.

   Locks: assumes we hold *all* the capabilities.
   -------------------------------------------------------------------------- */

void
resurrectThreads (StgTSO *threads)
{
    StgTSO *tso, *next;
    Capability *cap;
    generation *gen;

    for (tso = threads; tso != END_TSO_QUEUE; tso = next) {
        next = tso->global_link;

        gen = Bdescr((P_)tso)->gen;
        tso->global_link = gen->threads;
        gen->threads = tso;

        debugTrace(DEBUG_sched, "resurrecting thread %" FMT_StgThreadID, tso->id);

        // Wake up the thread on the Capability it was last on
        cap = tso->cap;

        switch (tso->why_blocked) {
        case BlockedOnMVar:
        case BlockedOnMVarRead:
            /* Called by GC - sched_mutex lock is currently held. */
            throwToSingleThreaded(cap, tso,
                                  (StgClosure *)blockedIndefinitelyOnMVar_closure);
            break;
        case BlockedOnBlackHole:
            throwToSingleThreaded(cap, tso,
                                  (StgClosure *)nonTermination_closure);
            break;
        case BlockedOnSTM:
            throwToSingleThreaded(cap, tso,
                                  (StgClosure *)blockedIndefinitelyOnSTM_closure);
            break;
        case NotBlocked:
            /* This might happen if the thread was blocked on a black hole
             * belonging to a thread that we've just woken up (raiseAsync
             * can wake up threads, remember...).
             */
            continue;
        case BlockedOnMsgThrowTo:
            // This can happen if the target is masking, blocks on a
            // black hole, and then is found to be unreachable.  In
            // this case, we want to let the target wake up and carry
            // on, and do nothing to this thread.
            continue;
        default:
            barf("resurrectThreads: thread blocked in a strange way: %d",
                 tso->why_blocked);
        }
    }
}