aboutsummaryrefslogtreecommitdiff
path: root/drivers/net/ethernet/tile/tilepro.c
blob: d9951afb9269ff130310dc912293f39d3808c666 (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
/*
 * Copyright 2011 Tilera Corporation. All Rights Reserved.
 *
 *   This program is free software; you can redistribute it and/or
 *   modify it under the terms of the GNU General Public License
 *   as published by the Free Software Foundation, version 2.
 *
 *   This program is distributed in the hope that it will be useful, but
 *   WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
 *   NON INFRINGEMENT.  See the GNU General Public License for
 *   more details.
 */

#include <linux/module.h>
#include <linux/init.h>
#include <linux/moduleparam.h>
#include <linux/sched.h>
#include <linux/kernel.h>      /* printk() */
#include <linux/slab.h>        /* kmalloc() */
#include <linux/errno.h>       /* error codes */
#include <linux/types.h>       /* size_t */
#include <linux/interrupt.h>
#include <linux/in.h>
#include <linux/netdevice.h>   /* struct device, and other headers */
#include <linux/etherdevice.h> /* eth_type_trans */
#include <linux/skbuff.h>
#include <linux/ioctl.h>
#include <linux/cdev.h>
#include <linux/hugetlb.h>
#include <linux/in6.h>
#include <linux/timer.h>
#include <linux/io.h>
#include <asm/checksum.h>
#include <asm/homecache.h>

#include <hv/drv_xgbe_intf.h>
#include <hv/drv_xgbe_impl.h>
#include <hv/hypervisor.h>
#include <hv/netio_intf.h>

/* For TSO */
#include <linux/ip.h>
#include <linux/tcp.h>


/*
 * First, "tile_net_init_module()" initializes all four "devices" which
 * can be used by linux.
 *
 * Then, "ifconfig DEVICE up" calls "tile_net_open()", which analyzes
 * the network cpus, then uses "tile_net_open_aux()" to initialize
 * LIPP/LEPP, and then uses "tile_net_open_inner()" to register all
 * the tiles, provide buffers to LIPP, allow ingress to start, and
 * turn on hypervisor interrupt handling (and NAPI) on all tiles.
 *
 * If registration fails due to the link being down, then "retry_work"
 * is used to keep calling "tile_net_open_inner()" until it succeeds.
 *
 * If "ifconfig DEVICE down" is called, it uses "tile_net_stop()" to
 * stop egress, drain the LIPP buffers, unregister all the tiles, stop
 * LIPP/LEPP, and wipe the LEPP queue.
 *
 * We start out with the ingress interrupt enabled on each CPU.  When
 * this interrupt fires, we disable it, and call "napi_schedule()".
 * This will cause "tile_net_poll()" to be called, which will pull
 * packets from the netio queue, filtering them out, or passing them
 * to "netif_receive_skb()".  If our budget is exhausted, we will
 * return, knowing we will be called again later.  Otherwise, we
 * reenable the ingress interrupt, and call "napi_complete()".
 *
 * HACK: Since disabling the ingress interrupt is not reliable, we
 * ignore the interrupt if the global "active" flag is false.
 *
 *
 * NOTE: The use of "native_driver" ensures that EPP exists, and that
 * we are using "LIPP" and "LEPP".
 *
 * NOTE: Failing to free completions for an arbitrarily long time
 * (which is defined to be illegal) does in fact cause bizarre
 * problems.  The "egress_timer" helps prevent this from happening.
 */


/* HACK: Allow use of "jumbo" packets. */
/* This should be 1500 if "jumbo" is not set in LIPP. */
/* This should be at most 10226 (10240 - 14) if "jumbo" is set in LIPP. */
/* ISSUE: This has not been thoroughly tested (except at 1500). */
#define TILE_NET_MTU 1500

/* HACK: Define to support GSO. */
/* ISSUE: This may actually hurt performance of the TCP blaster. */
/* #define TILE_NET_GSO */

/* Define this to collapse "duplicate" acks. */
/* #define IGNORE_DUP_ACKS */

/* HACK: Define this to verify incoming packets. */
/* #define TILE_NET_VERIFY_INGRESS */

/* Use 3000 to enable the Linux Traffic Control (QoS) layer, else 0. */
#define TILE_NET_TX_QUEUE_LEN 0

/* Define to dump packets (prints out the whole packet on tx and rx). */
/* #define TILE_NET_DUMP_PACKETS */

/* Define to enable debug spew (all PDEBUG's are enabled). */
/* #define TILE_NET_DEBUG */


/* Define to activate paranoia checks. */
/* #define TILE_NET_PARANOIA */

/* Default transmit lockup timeout period, in jiffies. */
#define TILE_NET_TIMEOUT (5 * HZ)

/* Default retry interval for bringing up the NetIO interface, in jiffies. */
#define TILE_NET_RETRY_INTERVAL (5 * HZ)

/* Number of ports (xgbe0, xgbe1, gbe0, gbe1). */
#define TILE_NET_DEVS 4



/* Paranoia. */
#if NET_IP_ALIGN != LIPP_PACKET_PADDING
#error "NET_IP_ALIGN must match LIPP_PACKET_PADDING."
#endif


/* Debug print. */
#ifdef TILE_NET_DEBUG
#define PDEBUG(fmt, args...) net_printk(fmt, ## args)
#else
#define PDEBUG(fmt, args...)
#endif


MODULE_AUTHOR("Tilera");
MODULE_LICENSE("GPL");


/*
 * Queue of incoming packets for a specific cpu and device.
 *
 * Includes a pointer to the "system" data, and the actual "user" data.
 */
struct tile_netio_queue {
	netio_queue_impl_t *__system_part;
	netio_queue_user_impl_t __user_part;

};


/*
 * Statistics counters for a specific cpu and device.
 */
struct tile_net_stats_t {
	u32 rx_packets;
	u32 rx_bytes;
	u32 tx_packets;
	u32 tx_bytes;
};


/*
 * Info for a specific cpu and device.
 *
 * ISSUE: There is a "dev" pointer in "napi" as well.
 */
struct tile_net_cpu {
	/* The NAPI struct. */
	struct napi_struct napi;
	/* Packet queue. */
	struct tile_netio_queue queue;
	/* Statistics. */
	struct tile_net_stats_t stats;
	/* True iff NAPI is enabled. */
	bool napi_enabled;
	/* True if this tile has successfully registered with the IPP. */
	bool registered;
	/* True if the link was down last time we tried to register. */
	bool link_down;
	/* True if "egress_timer" is scheduled. */
	bool egress_timer_scheduled;
	/* Number of small sk_buffs which must still be provided. */
	unsigned int num_needed_small_buffers;
	/* Number of large sk_buffs which must still be provided. */
	unsigned int num_needed_large_buffers;
	/* A timer for handling egress completions. */
	struct timer_list egress_timer;
};


/*
 * Info for a specific device.
 */
struct tile_net_priv {
	/* Our network device. */
	struct net_device *dev;
	/* Pages making up the egress queue. */
	struct page *eq_pages;
	/* Address of the actual egress queue. */
	lepp_queue_t *eq;
	/* Protects "eq". */
	spinlock_t eq_lock;
	/* The hypervisor handle for this interface. */
	int hv_devhdl;
	/* The intr bit mask that IDs this device. */
	u32 intr_id;
	/* True iff "tile_net_open_aux()" has succeeded. */
	bool partly_opened;
	/* True iff the device is "active". */
	bool active;
	/* Effective network cpus. */
	struct cpumask network_cpus_map;
	/* Number of network cpus. */
	int network_cpus_count;
	/* Credits per network cpu. */
	int network_cpus_credits;
	/* Network stats. */
	struct net_device_stats stats;
	/* For NetIO bringup retries. */
	struct delayed_work retry_work;
	/* Quick access to per cpu data. */
	struct tile_net_cpu *cpu[NR_CPUS];
};

/* Log2 of the number of small pages needed for the egress queue. */
#define EQ_ORDER  get_order(sizeof(lepp_queue_t))
/* Size of the egress queue's pages. */
#define EQ_SIZE   (1 << (PAGE_SHIFT + EQ_ORDER))

/*
 * The actual devices (xgbe0, xgbe1, gbe0, gbe1).
 */
static struct net_device *tile_net_devs[TILE_NET_DEVS];

/*
 * The "tile_net_cpu" structures for each device.
 */
static DEFINE_PER_CPU(struct tile_net_cpu, hv_xgbe0);
static DEFINE_PER_CPU(struct tile_net_cpu, hv_xgbe1);
static DEFINE_PER_CPU(struct tile_net_cpu, hv_gbe0);
static DEFINE_PER_CPU(struct tile_net_cpu, hv_gbe1);


/*
 * True if "network_cpus" was specified.
 */
static bool network_cpus_used;

/*
 * The actual cpus in "network_cpus".
 */
static struct cpumask network_cpus_map;



#ifdef TILE_NET_DEBUG
/*
 * printk with extra stuff.
 *
 * We print the CPU we're running in brackets.
 */
static void net_printk(char *fmt, ...)
{
	int i;
	int len;
	va_list args;
	static char buf[256];

	len = sprintf(buf, "tile_net[%2.2d]: ", smp_processor_id());
	va_start(args, fmt);
	i = vscnprintf(buf + len, sizeof(buf) - len - 1, fmt, args);
	va_end(args);
	buf[255] = '\0';
	pr_notice(buf);
}
#endif


#ifdef TILE_NET_DUMP_PACKETS
/*
 * Dump a packet.
 */
static void dump_packet(unsigned char *data, unsigned long length, char *s)
{
	int my_cpu = smp_processor_id();

	unsigned long i;
	char buf[128];

	static unsigned int count;

	pr_info("dump_packet(data %p, length 0x%lx s %s count 0x%x)\n",
	       data, length, s, count++);

	pr_info("\n");

	for (i = 0; i < length; i++) {
		if ((i & 0xf) == 0)
			sprintf(buf, "[%02d] %8.8lx:", my_cpu, i);
		sprintf(buf + strlen(buf), " %2.2x", data[i]);
		if ((i & 0xf) == 0xf || i == length - 1) {
			strcat(buf, "\n");
			pr_info("%s", buf);
		}
	}
}
#endif


/*
 * Provide support for the __netio_fastio1() swint
 * (see <hv/drv_xgbe_intf.h> for how it is used).
 *
 * The fastio swint2 call may clobber all the caller-saved registers.
 * It rarely clobbers memory, but we allow for the possibility in
 * the signature just to be on the safe side.
 *
 * Also, gcc doesn't seem to allow an input operand to be
 * clobbered, so we fake it with dummy outputs.
 *
 * This function can't be static because of the way it is declared
 * in the netio header.
 */
inline int __netio_fastio1(u32 fastio_index, u32 arg0)
{
	long result, clobber_r1, clobber_r10;
	asm volatile("swint2"
		     : "=R00" (result),
		       "=R01" (clobber_r1), "=R10" (clobber_r10)
		     : "R10" (fastio_index), "R01" (arg0)
		     : "memory", "r2", "r3", "r4",
		       "r5", "r6", "r7", "r8", "r9",
		       "r11", "r12", "r13", "r14",
		       "r15", "r16", "r17", "r18", "r19",
		       "r20", "r21", "r22", "r23", "r24",
		       "r25", "r26", "r27", "r28", "r29");
	return result;
}


/*
 * Provide a linux buffer to LIPP.
 */
static void tile_net_provide_linux_buffer(struct tile_net_cpu *info,
					  void *va, bool small)
{
	struct tile_netio_queue *queue = &info->queue;

	/* Convert "va" and "small" to "linux_buffer_t". */
	unsigned int buffer = ((unsigned int)(__pa(va) >> 7) << 1) + small;

	__netio_fastio_free_buffer(queue->__user_part.__fastio_index, buffer);
}


/*
 * Provide a linux buffer for LIPP.
 *
 * Note that the ACTUAL allocation for each buffer is a "struct sk_buff",
 * plus a chunk of memory that includes not only the requested bytes, but
 * also NET_SKB_PAD bytes of initial padding, and a "struct skb_shared_info".
 *
 * Note that "struct skb_shared_info" is 88 bytes with 64K pages and
 * 268 bytes with 4K pages (since the frags[] array needs 18 entries).
 *
 * Without jumbo packets, the maximum packet size will be 1536 bytes,
 * and we use 2 bytes (NET_IP_ALIGN) of padding.  ISSUE: If we told
 * the hardware to clip at 1518 bytes instead of 1536 bytes, then we
 * could save an entire cache line, but in practice, we don't need it.
 *
 * Since CPAs are 38 bits, and we can only encode the high 31 bits in
 * a "linux_buffer_t", the low 7 bits must be zero, and thus, we must
 * align the actual "va" mod 128.
 *
 * We assume that the underlying "head" will be aligned mod 64.  Note
 * that in practice, we have seen "head" NOT aligned mod 128 even when
 * using 2048 byte allocations, which is surprising.
 *
 * If "head" WAS always aligned mod 128, we could change LIPP to
 * assume that the low SIX bits are zero, and the 7th bit is one, that
 * is, align the actual "va" mod 128 plus 64, which would be "free".
 *
 * For now, the actual "head" pointer points at NET_SKB_PAD bytes of
 * padding, plus 28 or 92 bytes of extra padding, plus the sk_buff
 * pointer, plus the NET_IP_ALIGN padding, plus 126 or 1536 bytes for
 * the actual packet, plus 62 bytes of empty padding, plus some
 * padding and the "struct skb_shared_info".
 *
 * With 64K pages, a large buffer thus needs 32+92+4+2+1536+62+88
 * bytes, or 1816 bytes, which fits comfortably into 2048 bytes.
 *
 * With 64K pages, a small buffer thus needs 32+92+4+2+126+88
 * bytes, or 344 bytes, which means we are wasting 64+ bytes, and
 * could presumably increase the size of small buffers.
 *
 * With 4K pages, a large buffer thus needs 32+92+4+2+1536+62+268
 * bytes, or 1996 bytes, which fits comfortably into 2048 bytes.
 *
 * With 4K pages, a small buffer thus needs 32+92+4+2+126+268
 * bytes, or 524 bytes, which is annoyingly wasteful.
 *
 * Maybe we should increase LIPP_SMALL_PACKET_SIZE to 192?
 *
 * ISSUE: Maybe we should increase "NET_SKB_PAD" to 64?
 */
static bool tile_net_provide_needed_buffer(struct tile_net_cpu *info,
					   bool small)
{
#if TILE_NET_MTU <= 1536
	/* Without "jumbo", 2 + 1536 should be sufficient. */
	unsigned int large_size = NET_IP_ALIGN + 1536;
#else
	/* ISSUE: This has not been tested. */
	unsigned int large_size = NET_IP_ALIGN + TILE_NET_MTU + 100;
#endif

	/* Avoid "false sharing" with last cache line. */
	/* ISSUE: This is already done by "dev_alloc_skb()". */
	unsigned int len =
		 (((small ? LIPP_SMALL_PACKET_SIZE : large_size) +
		   CHIP_L2_LINE_SIZE() - 1) & -CHIP_L2_LINE_SIZE());

	unsigned int padding = 128 - NET_SKB_PAD;
	unsigned int align;

	struct sk_buff *skb;
	void *va;

	struct sk_buff **skb_ptr;

	/* Request 96 extra bytes for alignment purposes. */
	skb = dev_alloc_skb(len + padding);
	if (skb == NULL)
		return false;

	/* Skip 32 or 96 bytes to align "data" mod 128. */
	align = -(long)skb->data & (128 - 1);
	BUG_ON(align > padding);
	skb_reserve(skb, align);

	/* This address is given to IPP. */
	va = skb->data;

	/* Buffers must not span a huge page. */
	BUG_ON(((((long)va & ~HPAGE_MASK) + len) & HPAGE_MASK) != 0);

#ifdef TILE_NET_PARANOIA
#if CHIP_HAS_CBOX_HOME_MAP()
	if (hash_default) {
		HV_PTE pte = *virt_to_pte(current->mm, (unsigned long)va);
		if (hv_pte_get_mode(pte) != HV_PTE_MODE_CACHE_HASH_L3)
			panic("Non-HFH ingress buffer! VA=%p Mode=%d PTE=%llx",
			      va, hv_pte_get_mode(pte), hv_pte_val(pte));
	}
#endif
#endif

	/* Invalidate the packet buffer. */
	if (!hash_default)
		__inv_buffer(va, len);

	/* Skip two bytes to satisfy LIPP assumptions. */
	/* Note that this aligns IP on a 16 byte boundary. */
	/* ISSUE: Do this when the packet arrives? */
	skb_reserve(skb, NET_IP_ALIGN);

	/* Save a back-pointer to 'skb'. */
	skb_ptr = va - sizeof(*skb_ptr);
	*skb_ptr = skb;

	/* Make sure "skb_ptr" has been flushed. */
	__insn_mf();

	/* Provide the new buffer. */
	tile_net_provide_linux_buffer(info, va, small);

	return true;
}


/*
 * Provide linux buffers for LIPP.
 */
static void tile_net_provide_needed_buffers(struct tile_net_cpu *info)
{
	while (info->num_needed_small_buffers != 0) {
		if (!tile_net_provide_needed_buffer(info, true))
			goto oops;
		info->num_needed_small_buffers--;
	}

	while (info->num_needed_large_buffers != 0) {
		if (!tile_net_provide_needed_buffer(info, false))
			goto oops;
		info->num_needed_large_buffers--;
	}

	return;

oops:

	/* Add a description to the page allocation failure dump. */
	pr_notice("Could not provide a linux buffer to LIPP.\n");
}


/*
 * Grab some LEPP completions, and store them in "comps", of size
 * "comps_size", and return the number of completions which were
 * stored, so the caller can free them.
 */
static unsigned int tile_net_lepp_grab_comps(lepp_queue_t *eq,
					     struct sk_buff *comps[],
					     unsigned int comps_size,
					     unsigned int min_size)
{
	unsigned int n = 0;

	unsigned int comp_head = eq->comp_head;
	unsigned int comp_busy = eq->comp_busy;

	while (comp_head != comp_busy && n < comps_size) {
		comps[n++] = eq->comps[comp_head];
		LEPP_QINC(comp_head);
	}

	if (n < min_size)
		return 0;

	eq->comp_head = comp_head;

	return n;
}


/*
 * Free some comps, and return true iff there are still some pending.
 */
static bool tile_net_lepp_free_comps(struct net_device *dev, bool all)
{
	struct tile_net_priv *priv = netdev_priv(dev);

	lepp_queue_t *eq = priv->eq;

	struct sk_buff *olds[64];
	unsigned int wanted = 64;
	unsigned int i, n;
	bool pending;

	spin_lock(&priv->eq_lock);

	if (all)
		eq->comp_busy = eq->comp_tail;

	n = tile_net_lepp_grab_comps(eq, olds, wanted, 0);

	pending = (eq->comp_head != eq->comp_tail);

	spin_unlock(&priv->eq_lock);

	for (i = 0; i < n; i++)
		kfree_skb(olds[i]);

	return pending;
}


/*
 * Make sure the egress timer is scheduled.
 *
 * Note that we use "schedule if not scheduled" logic instead of the more
 * obvious "reschedule" logic, because "reschedule" is fairly expensive.
 */
static void tile_net_schedule_egress_timer(struct tile_net_cpu *info)
{
	if (!info->egress_timer_scheduled) {
		mod_timer_pinned(&info->egress_timer, jiffies + 1);
		info->egress_timer_scheduled = true;
	}
}


/*
 * The "function" for "info->egress_timer".
 *
 * This timer will reschedule itself as long as there are any pending
 * completions expected (on behalf of any tile).
 *
 * ISSUE: Realistically, will the timer ever stop scheduling itself?
 *
 * ISSUE: This timer is almost never actually needed, so just use a global
 * timer that can run on any tile.
 *
 * ISSUE: Maybe instead track number of expected completions, and free
 * only that many, resetting to zero if "pending" is ever false.
 */
static void tile_net_handle_egress_timer(unsigned long arg)
{
	struct tile_net_cpu *info = (struct tile_net_cpu *)arg;
	struct net_device *dev = info->napi.dev;

	/* The timer is no longer scheduled. */
	info->egress_timer_scheduled = false;

	/* Free comps, and reschedule timer if more are pending. */
	if (tile_net_lepp_free_comps(dev, false))
		tile_net_schedule_egress_timer(info);
}


#ifdef IGNORE_DUP_ACKS

/*
 * Help detect "duplicate" ACKs.  These are sequential packets (for a
 * given flow) which are exactly 66 bytes long, sharing everything but
 * ID=2@0x12, Hsum=2@0x18, Ack=4@0x2a, WinSize=2@0x30, Csum=2@0x32,
 * Tstamps=10@0x38.  The ID's are +1, the Hsum's are -1, the Ack's are
 * +N, and the Tstamps are usually identical.
 *
 * NOTE: Apparently truly duplicate acks (with identical "ack" values),
 * should not be collapsed, as they are used for some kind of flow control.
 */
static bool is_dup_ack(char *s1, char *s2, unsigned int len)
{
	int i;

	unsigned long long ignorable = 0;

	/* Identification. */
	ignorable |= (1ULL << 0x12);
	ignorable |= (1ULL << 0x13);

	/* Header checksum. */
	ignorable |= (1ULL << 0x18);
	ignorable |= (1ULL << 0x19);

	/* ACK. */
	ignorable |= (1ULL << 0x2a);
	ignorable |= (1ULL << 0x2b);
	ignorable |= (1ULL << 0x2c);
	ignorable |= (1ULL << 0x2d);

	/* WinSize. */
	ignorable |= (1ULL << 0x30);
	ignorable |= (1ULL << 0x31);

	/* Checksum. */
	ignorable |= (1ULL << 0x32);
	ignorable |= (1ULL << 0x33);

	for (i = 0; i < len; i++, ignorable >>= 1) {

		if ((ignorable & 1) || (s1[i] == s2[i]))
			continue;

#ifdef TILE_NET_DEBUG
		/* HACK: Mention non-timestamp diffs. */
		if (i < 0x38 && i != 0x2f &&
		    net_ratelimit())
			pr_info("Diff at 0x%x\n", i);
#endif

		return false;
	}

#ifdef TILE_NET_NO_SUPPRESS_DUP_ACKS
	/* HACK: Do not suppress truly duplicate ACKs. */
	/* ISSUE: Is this actually necessary or helpful? */
	if (s1[0x2a] == s2[0x2a] &&
	    s1[0x2b] == s2[0x2b] &&
	    s1[0x2c] == s2[0x2c] &&
	    s1[0x2d] == s2[0x2d]) {
		return false;
	}
#endif

	return true;
}

#endif



static void tile_net_discard_aux(struct tile_net_cpu *info, int index)
{
	struct tile_netio_queue *queue = &info->queue;
	netio_queue_impl_t *qsp = queue->__system_part;
	netio_queue_user_impl_t *qup = &queue->__user_part;

	int index2_aux = index + sizeof(netio_pkt_t);
	int index2 =
		((index2_aux ==
		  qsp->__packet_receive_queue.__last_packet_plus_one) ?
		 0 : index2_aux);

	netio_pkt_t *pkt = (netio_pkt_t *)((unsigned long) &qsp[1] + index);

	/* Extract the "linux_buffer_t". */
	unsigned int buffer = pkt->__packet.word;

	/* Convert "linux_buffer_t" to "va". */
	void *va = __va((phys_addr_t)(buffer >> 1) << 7);

	/* Acquire the associated "skb". */
	struct sk_buff **skb_ptr = va - sizeof(*skb_ptr);
	struct sk_buff *skb = *skb_ptr;

	kfree_skb(skb);

	/* Consume this packet. */
	qup->__packet_receive_read = index2;
}


/*
 * Like "tile_net_poll()", but just discard packets.
 */
static void tile_net_discard_packets(struct net_device *dev)
{
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info = priv->cpu[my_cpu];
	struct tile_netio_queue *queue = &info->queue;
	netio_queue_impl_t *qsp = queue->__system_part;
	netio_queue_user_impl_t *qup = &queue->__user_part;

	while (qup->__packet_receive_read !=
	       qsp->__packet_receive_queue.__packet_write) {
		int index = qup->__packet_receive_read;
		tile_net_discard_aux(info, index);
	}
}


/*
 * Handle the next packet.  Return true if "processed", false if "filtered".
 */
static bool tile_net_poll_aux(struct tile_net_cpu *info, int index)
{
	struct net_device *dev = info->napi.dev;

	struct tile_netio_queue *queue = &info->queue;
	netio_queue_impl_t *qsp = queue->__system_part;
	netio_queue_user_impl_t *qup = &queue->__user_part;
	struct tile_net_stats_t *stats = &info->stats;

	int filter;

	int index2_aux = index + sizeof(netio_pkt_t);
	int index2 =
		((index2_aux ==
		  qsp->__packet_receive_queue.__last_packet_plus_one) ?
		 0 : index2_aux);

	netio_pkt_t *pkt = (netio_pkt_t *)((unsigned long) &qsp[1] + index);

	netio_pkt_metadata_t *metadata = NETIO_PKT_METADATA(pkt);

	/* Extract the packet size.  FIXME: Shouldn't the second line */
	/* get subtracted?  Mostly moot, since it should be "zero". */
	unsigned long len =
		(NETIO_PKT_CUSTOM_LENGTH(pkt) +
		 NET_IP_ALIGN - NETIO_PACKET_PADDING);

	/* Extract the "linux_buffer_t". */
	unsigned int buffer = pkt->__packet.word;

	/* Extract "small" (vs "large"). */
	bool small = ((buffer & 1) != 0);

	/* Convert "linux_buffer_t" to "va". */
	void *va = __va((phys_addr_t)(buffer >> 1) << 7);

	/* Extract the packet data pointer. */
	/* Compare to "NETIO_PKT_CUSTOM_DATA(pkt)". */
	unsigned char *buf = va + NET_IP_ALIGN;

	/* Invalidate the packet buffer. */
	if (!hash_default)
		__inv_buffer(buf, len);

	/* ISSUE: Is this needed? */
	dev->last_rx = jiffies;

#ifdef TILE_NET_DUMP_PACKETS
	dump_packet(buf, len, "rx");
#endif /* TILE_NET_DUMP_PACKETS */

#ifdef TILE_NET_VERIFY_INGRESS
	if (!NETIO_PKT_L4_CSUM_CORRECT_M(metadata, pkt) &&
	    NETIO_PKT_L4_CSUM_CALCULATED_M(metadata, pkt)) {
		/* Bug 6624: Includes UDP packets with a "zero" checksum. */
		pr_warning("Bad L4 checksum on %d byte packet.\n", len);
	}
	if (!NETIO_PKT_L3_CSUM_CORRECT_M(metadata, pkt) &&
	    NETIO_PKT_L3_CSUM_CALCULATED_M(metadata, pkt)) {
		dump_packet(buf, len, "rx");
		panic("Bad L3 checksum.");
	}
	switch (NETIO_PKT_STATUS_M(metadata, pkt)) {
	case NETIO_PKT_STATUS_OVERSIZE:
		if (len >= 64) {
			dump_packet(buf, len, "rx");
			panic("Unexpected OVERSIZE.");
		}
		break;
	case NETIO_PKT_STATUS_BAD:
		pr_warning("Unexpected BAD %ld byte packet.\n", len);
	}
#endif

	filter = 0;

	/* ISSUE: Filter TCP packets with "bad" checksums? */

	if (!(dev->flags & IFF_UP)) {
		/* Filter packets received before we're up. */
		filter = 1;
	} else if (NETIO_PKT_STATUS_M(metadata, pkt) == NETIO_PKT_STATUS_BAD) {
		/* Filter "truncated" packets. */
		filter = 1;
	} else if (!(dev->flags & IFF_PROMISC)) {
		/* FIXME: Implement HW multicast filter. */
		if (!is_multicast_ether_addr(buf)) {
			/* Filter packets not for our address. */
			const u8 *mine = dev->dev_addr;
			filter = compare_ether_addr(mine, buf);
		}
	}

	if (filter) {

		/* ISSUE: Update "drop" statistics? */

		tile_net_provide_linux_buffer(info, va, small);

	} else {

		/* Acquire the associated "skb". */
		struct sk_buff **skb_ptr = va - sizeof(*skb_ptr);
		struct sk_buff *skb = *skb_ptr;

		/* Paranoia. */
		if (skb->data != buf)
			panic("Corrupt linux buffer from LIPP! "
			      "VA=%p, skb=%p, skb->data=%p\n",
			      va, skb, skb->data);

		/* Encode the actual packet length. */
		skb_put(skb, len);

		/* NOTE: This call also sets "skb->dev = dev". */
		skb->protocol = eth_type_trans(skb, dev);

		/* Avoid recomputing "good" TCP/UDP checksums. */
		if (NETIO_PKT_L4_CSUM_CORRECT_M(metadata, pkt))
			skb->ip_summed = CHECKSUM_UNNECESSARY;

		netif_receive_skb(skb);

		stats->rx_packets++;
		stats->rx_bytes += len;

		if (small)
			info->num_needed_small_buffers++;
		else
			info->num_needed_large_buffers++;
	}

	/* Return four credits after every fourth packet. */
	if (--qup->__receive_credit_remaining == 0) {
		u32 interval = qup->__receive_credit_interval;
		qup->__receive_credit_remaining = interval;
		__netio_fastio_return_credits(qup->__fastio_index, interval);
	}

	/* Consume this packet. */
	qup->__packet_receive_read = index2;

	return !filter;
}


/*
 * Handle some packets for the given device on the current CPU.
 *
 * If "tile_net_stop()" is called on some other tile while this
 * function is running, we will return, hopefully before that
 * other tile asks us to call "napi_disable()".
 *
 * The "rotting packet" race condition occurs if a packet arrives
 * during the extremely narrow window between the queue appearing to
 * be empty, and the ingress interrupt being re-enabled.  This happens
 * a LOT under heavy network load.
 */
static int tile_net_poll(struct napi_struct *napi, int budget)
{
	struct net_device *dev = napi->dev;
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info = priv->cpu[my_cpu];
	struct tile_netio_queue *queue = &info->queue;
	netio_queue_impl_t *qsp = queue->__system_part;
	netio_queue_user_impl_t *qup = &queue->__user_part;

	unsigned int work = 0;

	while (priv->active) {
		int index = qup->__packet_receive_read;
		if (index == qsp->__packet_receive_queue.__packet_write)
			break;

		if (tile_net_poll_aux(info, index)) {
			if (++work >= budget)
				goto done;
		}
	}

	napi_complete(&info->napi);

	if (!priv->active)
		goto done;

	/* Re-enable the ingress interrupt. */
	enable_percpu_irq(priv->intr_id, 0);

	/* HACK: Avoid the "rotting packet" problem (see above). */
	if (qup->__packet_receive_read !=
	    qsp->__packet_receive_queue.__packet_write) {
		/* ISSUE: Sometimes this returns zero, presumably */
		/* because an interrupt was handled for this tile. */
		(void)napi_reschedule(&info->napi);
	}

done:

	if (priv->active)
		tile_net_provide_needed_buffers(info);

	return work;
}


/*
 * Handle an ingress interrupt for the given device on the current cpu.
 *
 * ISSUE: Sometimes this gets called after "disable_percpu_irq()" has
 * been called!  This is probably due to "pending hypervisor downcalls".
 *
 * ISSUE: Is there any race condition between the "napi_schedule()" here
 * and the "napi_complete()" call above?
 */
static irqreturn_t tile_net_handle_ingress_interrupt(int irq, void *dev_ptr)
{
	struct net_device *dev = (struct net_device *)dev_ptr;
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info = priv->cpu[my_cpu];

	/* Disable the ingress interrupt. */
	disable_percpu_irq(priv->intr_id);

	/* Ignore unwanted interrupts. */
	if (!priv->active)
		return IRQ_HANDLED;

	/* ISSUE: Sometimes "info->napi_enabled" is false here. */

	napi_schedule(&info->napi);

	return IRQ_HANDLED;
}


/*
 * One time initialization per interface.
 */
static int tile_net_open_aux(struct net_device *dev)
{
	struct tile_net_priv *priv = netdev_priv(dev);

	int ret;
	int dummy;
	unsigned int epp_lotar;

	/*
	 * Find out where EPP memory should be homed.
	 */
	ret = hv_dev_pread(priv->hv_devhdl, 0,
			   (HV_VirtAddr)&epp_lotar, sizeof(epp_lotar),
			   NETIO_EPP_SHM_OFF);
	if (ret < 0) {
		pr_err("could not read epp_shm_queue lotar.\n");
		return -EIO;
	}

	/*
	 * Home the page on the EPP.
	 */
	{
		int epp_home = hv_lotar_to_cpu(epp_lotar);
		homecache_change_page_home(priv->eq_pages, EQ_ORDER, epp_home);
	}

	/*
	 * Register the EPP shared memory queue.
	 */
	{
		netio_ipp_address_t ea = {
			.va = 0,
			.pa = __pa(priv->eq),
			.pte = hv_pte(0),
			.size = EQ_SIZE,
		};
		ea.pte = hv_pte_set_lotar(ea.pte, epp_lotar);
		ea.pte = hv_pte_set_mode(ea.pte, HV_PTE_MODE_CACHE_TILE_L3);
		ret = hv_dev_pwrite(priv->hv_devhdl, 0,
				    (HV_VirtAddr)&ea,
				    sizeof(ea),
				    NETIO_EPP_SHM_OFF);
		if (ret < 0)
			return -EIO;
	}

	/*
	 * Start LIPP/LEPP.
	 */
	if (hv_dev_pwrite(priv->hv_devhdl, 0, (HV_VirtAddr)&dummy,
			  sizeof(dummy), NETIO_IPP_START_SHIM_OFF) < 0) {
		pr_warning("Failed to start LIPP/LEPP.\n");
		return -EIO;
	}

	return 0;
}


/*
 * Register with hypervisor on the current CPU.
 *
 * Strangely, this function does important things even if it "fails",
 * which is especially common if the link is not up yet.  Hopefully
 * these things are all "harmless" if done twice!
 */
static void tile_net_register(void *dev_ptr)
{
	struct net_device *dev = (struct net_device *)dev_ptr;
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info;

	struct tile_netio_queue *queue;

	/* Only network cpus can receive packets. */
	int queue_id =
		cpumask_test_cpu(my_cpu, &priv->network_cpus_map) ? 0 : 255;

	netio_input_config_t config = {
		.flags = 0,
		.num_receive_packets = priv->network_cpus_credits,
		.queue_id = queue_id
	};

	int ret = 0;
	netio_queue_impl_t *queuep;

	PDEBUG("tile_net_register(queue_id %d)\n", queue_id);

	if (!strcmp(dev->name, "xgbe0"))
		info = &__get_cpu_var(hv_xgbe0);
	else if (!strcmp(dev->name, "xgbe1"))
		info = &__get_cpu_var(hv_xgbe1);
	else if (!strcmp(dev->name, "gbe0"))
		info = &__get_cpu_var(hv_gbe0);
	else if (!strcmp(dev->name, "gbe1"))
		info = &__get_cpu_var(hv_gbe1);
	else
		BUG();

	/* Initialize the egress timer. */
	init_timer(&info->egress_timer);
	info->egress_timer.data = (long)info;
	info->egress_timer.function = tile_net_handle_egress_timer;

	priv->cpu[my_cpu] = info;

	/*
	 * Register ourselves with LIPP.  This does a lot of stuff,
	 * including invoking the LIPP registration code.
	 */
	ret = hv_dev_pwrite(priv->hv_devhdl, 0,
			    (HV_VirtAddr)&config,
			    sizeof(netio_input_config_t),
			    NETIO_IPP_INPUT_REGISTER_OFF);
	PDEBUG("hv_dev_pwrite(NETIO_IPP_INPUT_REGISTER_OFF) returned %d\n",
	       ret);
	if (ret < 0) {
		if (ret != NETIO_LINK_DOWN) {
			printk(KERN_DEBUG "hv_dev_pwrite "
			       "NETIO_IPP_INPUT_REGISTER_OFF failure %d\n",
			       ret);
		}
		info->link_down = (ret == NETIO_LINK_DOWN);
		return;
	}

	/*
	 * Get the pointer to our queue's system part.
	 */

	ret = hv_dev_pread(priv->hv_devhdl, 0,
			   (HV_VirtAddr)&queuep,
			   sizeof(netio_queue_impl_t *),
			   NETIO_IPP_INPUT_REGISTER_OFF);
	PDEBUG("hv_dev_pread(NETIO_IPP_INPUT_REGISTER_OFF) returned %d\n",
	       ret);
	PDEBUG("queuep %p\n", queuep);
	if (ret <= 0) {
		/* ISSUE: Shouldn't this be a fatal error? */
		pr_err("hv_dev_pread NETIO_IPP_INPUT_REGISTER_OFF failure\n");
		return;
	}

	queue = &info->queue;

	queue->__system_part = queuep;

	memset(&queue->__user_part, 0, sizeof(netio_queue_user_impl_t));

	/* This is traditionally "config.num_receive_packets / 2". */
	queue->__user_part.__receive_credit_interval = 4;
	queue->__user_part.__receive_credit_remaining =
		queue->__user_part.__receive_credit_interval;

	/*
	 * Get a fastio index from the hypervisor.
	 * ISSUE: Shouldn't this check the result?
	 */
	ret = hv_dev_pread(priv->hv_devhdl, 0,
			   (HV_VirtAddr)&queue->__user_part.__fastio_index,
			   sizeof(queue->__user_part.__fastio_index),
			   NETIO_IPP_GET_FASTIO_OFF);
	PDEBUG("hv_dev_pread(NETIO_IPP_GET_FASTIO_OFF) returned %d\n", ret);

	/* Now we are registered. */
	info->registered = true;
}


/*
 * Deregister with hypervisor on the current CPU.
 *
 * This simply discards all our credits, so no more packets will be
 * delivered to this tile.  There may still be packets in our queue.
 *
 * Also, disable the ingress interrupt.
 */
static void tile_net_deregister(void *dev_ptr)
{
	struct net_device *dev = (struct net_device *)dev_ptr;
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info = priv->cpu[my_cpu];

	/* Disable the ingress interrupt. */
	disable_percpu_irq(priv->intr_id);

	/* Do nothing else if not registered. */
	if (info == NULL || !info->registered)
		return;

	{
		struct tile_netio_queue *queue = &info->queue;
		netio_queue_user_impl_t *qup = &queue->__user_part;

		/* Discard all our credits. */
		__netio_fastio_return_credits(qup->__fastio_index, -1);
	}
}


/*
 * Unregister with hypervisor on the current CPU.
 *
 * Also, disable the ingress interrupt.
 */
static void tile_net_unregister(void *dev_ptr)
{
	struct net_device *dev = (struct net_device *)dev_ptr;
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info = priv->cpu[my_cpu];

	int ret;
	int dummy = 0;

	/* Disable the ingress interrupt. */
	disable_percpu_irq(priv->intr_id);

	/* Do nothing else if not registered. */
	if (info == NULL || !info->registered)
		return;

	/* Unregister ourselves with LIPP/LEPP. */
	ret = hv_dev_pwrite(priv->hv_devhdl, 0, (HV_VirtAddr)&dummy,
			    sizeof(dummy), NETIO_IPP_INPUT_UNREGISTER_OFF);
	if (ret < 0)
		panic("Failed to unregister with LIPP/LEPP!\n");

	/* Discard all packets still in our NetIO queue. */
	tile_net_discard_packets(dev);

	/* Reset state. */
	info->num_needed_small_buffers = 0;
	info->num_needed_large_buffers = 0;

	/* Cancel egress timer. */
	del_timer(&info->egress_timer);
	info->egress_timer_scheduled = false;
}


/*
 * Helper function for "tile_net_stop()".
 *
 * Also used to handle registration failure in "tile_net_open_inner()",
 * when the various extra steps in "tile_net_stop()" are not necessary.
 */
static void tile_net_stop_aux(struct net_device *dev)
{
	struct tile_net_priv *priv = netdev_priv(dev);
	int i;

	int dummy = 0;

	/*
	 * Unregister all tiles, so LIPP will stop delivering packets.
	 * Also, delete all the "napi" objects (sequentially, to protect
	 * "dev->napi_list").
	 */
	on_each_cpu(tile_net_unregister, (void *)dev, 1);
	for_each_online_cpu(i) {
		struct tile_net_cpu *info = priv->cpu[i];
		if (info != NULL && info->registered) {
			netif_napi_del(&info->napi);
			info->registered = false;
		}
	}

	/* Stop LIPP/LEPP. */
	if (hv_dev_pwrite(priv->hv_devhdl, 0, (HV_VirtAddr)&dummy,
			  sizeof(dummy), NETIO_IPP_STOP_SHIM_OFF) < 0)
		panic("Failed to stop LIPP/LEPP!\n");

	priv->partly_opened = false;
}


/*
 * Disable NAPI for the given device on the current cpu.
 */
static void tile_net_stop_disable(void *dev_ptr)
{
	struct net_device *dev = (struct net_device *)dev_ptr;
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info = priv->cpu[my_cpu];

	/* Disable NAPI if needed. */
	if (info != NULL && info->napi_enabled) {
		napi_disable(&info->napi);
		info->napi_enabled = false;
	}
}


/*
 * Enable NAPI and the ingress interrupt for the given device
 * on the current cpu.
 *
 * ISSUE: Only do this for "network cpus"?
 */
static void tile_net_open_enable(void *dev_ptr)
{
	struct net_device *dev = (struct net_device *)dev_ptr;
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info = priv->cpu[my_cpu];

	/* Enable NAPI. */
	napi_enable(&info->napi);
	info->napi_enabled = true;

	/* Enable the ingress interrupt. */
	enable_percpu_irq(priv->intr_id, 0);
}


/*
 * tile_net_open_inner does most of the work of bringing up the interface.
 * It's called from tile_net_open(), and also from tile_net_retry_open().
 * The return value is 0 if the interface was brought up, < 0 if
 * tile_net_open() should return the return value as an error, and > 0 if
 * tile_net_open() should return success and schedule a work item to
 * periodically retry the bringup.
 */
static int tile_net_open_inner(struct net_device *dev)
{
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info;
	struct tile_netio_queue *queue;
	int result = 0;
	int i;
	int dummy = 0;

	/*
	 * First try to register just on the local CPU, and handle any
	 * semi-expected "link down" failure specially.  Note that we
	 * do NOT call "tile_net_stop_aux()", unlike below.
	 */
	tile_net_register(dev);
	info = priv->cpu[my_cpu];
	if (!info->registered) {
		if (info->link_down)
			return 1;
		return -EAGAIN;
	}

	/*
	 * Now register everywhere else.  If any registration fails,
	 * even for "link down" (which might not be possible), we
	 * clean up using "tile_net_stop_aux()".  Also, add all the
	 * "napi" objects (sequentially, to protect "dev->napi_list").
	 * ISSUE: Only use "netif_napi_add()" for "network cpus"?
	 */
	smp_call_function(tile_net_register, (void *)dev, 1);
	for_each_online_cpu(i) {
		struct tile_net_cpu *info = priv->cpu[i];
		if (info->registered)
			netif_napi_add(dev, &info->napi, tile_net_poll, 64);
		else
			result = -EAGAIN;
	}
	if (result != 0) {
		tile_net_stop_aux(dev);
		return result;
	}

	queue = &info->queue;

	if (priv->intr_id == 0) {
		unsigned int irq;

		/*
		 * Acquire the irq allocated by the hypervisor.  Every
		 * queue gets the same irq.  The "__intr_id" field is
		 * "1 << irq", so we use "__ffs()" to extract "irq".
		 */
		priv->intr_id = queue->__system_part->__intr_id;
		BUG_ON(priv->intr_id == 0);
		irq = __ffs(priv->intr_id);

		/*
		 * Register the ingress interrupt handler for this
		 * device, permanently.
		 *
		 * We used to call "free_irq()" in "tile_net_stop()",
		 * and then re-register the handler here every time,
		 * but that caused DNP errors in "handle_IRQ_event()"
		 * because "desc->action" was NULL.  See bug 9143.
		 */
		tile_irq_activate(irq, TILE_IRQ_PERCPU);
		BUG_ON(request_irq(irq, tile_net_handle_ingress_interrupt,
				   0, dev->name, (void *)dev) != 0);
	}

	{
		/* Allocate initial buffers. */

		int max_buffers =
			priv->network_cpus_count * priv->network_cpus_credits;

		info->num_needed_small_buffers =
			min(LIPP_SMALL_BUFFERS, max_buffers);

		info->num_needed_large_buffers =
			min(LIPP_LARGE_BUFFERS, max_buffers);

		tile_net_provide_needed_buffers(info);

		if (info->num_needed_small_buffers != 0 ||
		    info->num_needed_large_buffers != 0)
			panic("Insufficient memory for buffer stack!");
	}

	/* We are about to be active. */
	priv->active = true;

	/* Make sure "active" is visible to all tiles. */
	mb();

	/* On each tile, enable NAPI and the ingress interrupt. */
	on_each_cpu(tile_net_open_enable, (void *)dev, 1);

	/* Start LIPP/LEPP and activate "ingress" at the shim. */
	if (hv_dev_pwrite(priv->hv_devhdl, 0, (HV_VirtAddr)&dummy,
			  sizeof(dummy), NETIO_IPP_INPUT_INIT_OFF) < 0)
		panic("Failed to activate the LIPP Shim!\n");

	/* Start our transmit queue. */
	netif_start_queue(dev);

	return 0;
}


/*
 * Called periodically to retry bringing up the NetIO interface,
 * if it doesn't come up cleanly during tile_net_open().
 */
static void tile_net_open_retry(struct work_struct *w)
{
	struct delayed_work *dw =
		container_of(w, struct delayed_work, work);

	struct tile_net_priv *priv =
		container_of(dw, struct tile_net_priv, retry_work);

	/*
	 * Try to bring the NetIO interface up.  If it fails, reschedule
	 * ourselves to try again later; otherwise, tell Linux we now have
	 * a working link.  ISSUE: What if the return value is negative?
	 */
	if (tile_net_open_inner(priv->dev) != 0)
		schedule_delayed_work(&priv->retry_work,
				      TILE_NET_RETRY_INTERVAL);
	else
		netif_carrier_on(priv->dev);
}


/*
 * Called when a network interface is made active.
 *
 * Returns 0 on success, negative value on failure.
 *
 * The open entry point is called when a network interface is made
 * active by the system (IFF_UP).  At this point all resources needed
 * for transmit and receive operations are allocated, the interrupt
 * handler is registered with the OS (if needed), the watchdog timer
 * is started, and the stack is notified that the interface is ready.
 *
 * If the actual link is not available yet, then we tell Linux that
 * we have no carrier, and we keep checking until the link comes up.
 */
static int tile_net_open(struct net_device *dev)
{
	int ret = 0;
	struct tile_net_priv *priv = netdev_priv(dev);

	/*
	 * We rely on priv->partly_opened to tell us if this is the
	 * first time this interface is being brought up. If it is
	 * set, the IPP was already initialized and should not be
	 * initialized again.
	 */
	if (!priv->partly_opened) {

		int count;
		int credits;

		/* Initialize LIPP/LEPP, and start the Shim. */
		ret = tile_net_open_aux(dev);
		if (ret < 0) {
			pr_err("tile_net_open_aux failed: %d\n", ret);
			return ret;
		}

		/* Analyze the network cpus. */

		if (network_cpus_used)
			cpumask_copy(&priv->network_cpus_map,
				     &network_cpus_map);
		else
			cpumask_copy(&priv->network_cpus_map, cpu_online_mask);


		count = cpumask_weight(&priv->network_cpus_map);

		/* Limit credits to available buffers, and apply min. */
		credits = max(16, (LIPP_LARGE_BUFFERS / count) & ~1);

		/* Apply "GBE" max limit. */
		/* ISSUE: Use higher limit for XGBE? */
		credits = min(NETIO_MAX_RECEIVE_PKTS, credits);

		priv->network_cpus_count = count;
		priv->network_cpus_credits = credits;

#ifdef TILE_NET_DEBUG
		pr_info("Using %d network cpus, with %d credits each\n",
		       priv->network_cpus_count, priv->network_cpus_credits);
#endif

		priv->partly_opened = true;

	} else {
		/* FIXME: Is this possible? */
		/* printk("Already partly opened.\n"); */
	}

	/*
	 * Attempt to bring up the link.
	 */
	ret = tile_net_open_inner(dev);
	if (ret <= 0) {
		if (ret == 0)
			netif_carrier_on(dev);
		return ret;
	}

	/*
	 * We were unable to bring up the NetIO interface, but we want to
	 * try again in a little bit.  Tell Linux that we have no carrier
	 * so it doesn't try to use the interface before the link comes up
	 * and then remember to try again later.
	 */
	netif_carrier_off(dev);
	schedule_delayed_work(&priv->retry_work, TILE_NET_RETRY_INTERVAL);

	return 0;
}


static int tile_net_drain_lipp_buffers(struct tile_net_priv *priv)
{
	int n = 0;

	/* Drain all the LIPP buffers. */
	while (true) {
		int buffer;

		/* NOTE: This should never fail. */
		if (hv_dev_pread(priv->hv_devhdl, 0, (HV_VirtAddr)&buffer,
				 sizeof(buffer), NETIO_IPP_DRAIN_OFF) < 0)
			break;

		/* Stop when done. */
		if (buffer == 0)
			break;

		{
			/* Convert "linux_buffer_t" to "va". */
			void *va = __va((phys_addr_t)(buffer >> 1) << 7);

			/* Acquire the associated "skb". */
			struct sk_buff **skb_ptr = va - sizeof(*skb_ptr);
			struct sk_buff *skb = *skb_ptr;

			kfree_skb(skb);
		}

		n++;
	}

	return n;
}


/*
 * Disables a network interface.
 *
 * Returns 0, this is not allowed to fail.
 *
 * The close entry point is called when an interface is de-activated
 * by the OS.  The hardware is still under the drivers control, but
 * needs to be disabled.  A global MAC reset is issued to stop the
 * hardware, and all transmit and receive resources are freed.
 *
 * ISSUE: How closely does "netif_running(dev)" mirror "priv->active"?
 *
 * Before we are called by "__dev_close()", "netif_running()" will
 * have been cleared, so no NEW calls to "tile_net_poll()" will be
 * made by "netpoll_poll_dev()".
 *
 * Often, this can cause some tiles to still have packets in their
 * queues, so we must call "tile_net_discard_packets()" later.
 *
 * Note that some other tile may still be INSIDE "tile_net_poll()",
 * and in fact, many will be, if there is heavy network load.
 *
 * Calling "on_each_cpu(tile_net_stop_disable, (void *)dev, 1)" when
 * any tile is still "napi_schedule()"'d will induce a horrible crash
 * when "msleep()" is called.  This includes tiles which are inside
 * "tile_net_poll()" which have not yet called "napi_complete()".
 *
 * So, we must first try to wait long enough for other tiles to finish
 * with any current "tile_net_poll()" call, and, hopefully, to clear
 * the "scheduled" flag.  ISSUE: It is unclear what happens to tiles
 * which have called "napi_schedule()" but which had not yet tried to
 * call "tile_net_poll()", or which exhausted their budget inside
 * "tile_net_poll()" just before this function was called.
 */
static int tile_net_stop(struct net_device *dev)
{
	struct tile_net_priv *priv = netdev_priv(dev);

	PDEBUG("tile_net_stop()\n");

	/* Start discarding packets. */
	priv->active = false;

	/* Make sure "active" is visible to all tiles. */
	mb();

	/*
	 * On each tile, make sure no NEW packets get delivered, and
	 * disable the ingress interrupt.
	 *
	 * Note that the ingress interrupt can fire AFTER this,
	 * presumably due to packets which were recently delivered,
	 * but it will have no effect.
	 */
	on_each_cpu(tile_net_deregister, (void *)dev, 1);

	/* Optimistically drain LIPP buffers. */
	(void)tile_net_drain_lipp_buffers(priv);

	/* ISSUE: Only needed if not yet fully open. */
	cancel_delayed_work_sync(&priv->retry_work);

	/* Can't transmit any more. */
	netif_stop_queue(dev);

	/* Disable NAPI on each tile. */
	on_each_cpu(tile_net_stop_disable, (void *)dev, 1);

	/*
	 * Drain any remaining LIPP buffers.  NOTE: This "printk()"
	 * has never been observed, but in theory it could happen.
	 */
	if (tile_net_drain_lipp_buffers(priv) != 0)
		printk("Had to drain some extra LIPP buffers!\n");

	/* Stop LIPP/LEPP. */
	tile_net_stop_aux(dev);

	/*
	 * ISSUE: It appears that, in practice anyway, by the time we
	 * get here, there are no pending completions, but just in case,
	 * we free (all of) them anyway.
	 */
	while (tile_net_lepp_free_comps(dev, true))
		/* loop */;

	/* Wipe the EPP queue, and wait till the stores hit the EPP. */
	memset(priv->eq, 0, sizeof(lepp_queue_t));
	mb();

	return 0;
}


/*
 * Prepare the "frags" info for the resulting LEPP command.
 *
 * If needed, flush the memory used by the frags.
 */
static unsigned int tile_net_tx_frags(lepp_frag_t *frags,
				      struct sk_buff *skb,
				      void *b_data, unsigned int b_len)
{
	unsigned int i, n = 0;

	struct skb_shared_info *sh = skb_shinfo(skb);

	phys_addr_t cpa;

	if (b_len != 0) {

		if (!hash_default)
			finv_buffer_remote(b_data, b_len, 0);

		cpa = __pa(b_data);
		frags[n].cpa_lo = cpa;
		frags[n].cpa_hi = cpa >> 32;
		frags[n].length = b_len;
		frags[n].hash_for_home = hash_default;
		n++;
	}

	for (i = 0; i < sh->nr_frags; i++) {

		skb_frag_t *f = &sh->frags[i];
		unsigned long pfn = page_to_pfn(skb_frag_page(f));

		/* FIXME: Compute "hash_for_home" properly. */
		/* ISSUE: The hypervisor checks CHIP_HAS_REV1_DMA_PACKETS(). */
		int hash_for_home = hash_default;

		/* FIXME: Hmmm. */
		if (!hash_default) {
			void *va = pfn_to_kaddr(pfn) + f->page_offset;
			BUG_ON(PageHighMem(skb_frag_page(f)));
			finv_buffer_remote(va, f->size, 0);
		}

		cpa = ((phys_addr_t)pfn << PAGE_SHIFT) + f->page_offset;
		frags[n].cpa_lo = cpa;
		frags[n].cpa_hi = cpa >> 32;
		frags[n].length = skb_frag_size(f);
		frags[n].hash_for_home = hash_for_home;
		n++;
	}

	return n;
}


/*
 * This function takes "skb", consisting of a header template and a
 * payload, and hands it to LEPP, to emit as one or more segments,
 * each consisting of a possibly modified header, plus a piece of the
 * payload, via a process known as "tcp segmentation offload".
 *
 * Usually, "data" will contain the header template, of size "sh_len",
 * and "sh->frags" will contain "skb->data_len" bytes of payload, and
 * there will be "sh->gso_segs" segments.
 *
 * Sometimes, if "sendfile()" requires copying, we will be called with
 * "data" containing the header and payload, with "frags" being empty.
 *
 * In theory, "sh->nr_frags" could be 3, but in practice, it seems
 * that this will never actually happen.
 *
 * See "emulate_large_send_offload()" for some reference code, which
 * does not handle checksumming.
 *
 * ISSUE: How do we make sure that high memory DMA does not migrate?
 */
static int tile_net_tx_tso(struct sk_buff *skb, struct net_device *dev)
{
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info = priv->cpu[my_cpu];
	struct tile_net_stats_t *stats = &info->stats;

	struct skb_shared_info *sh = skb_shinfo(skb);

	unsigned char *data = skb->data;

	/* The ip header follows the ethernet header. */
	struct iphdr *ih = ip_hdr(skb);
	unsigned int ih_len = ih->ihl * 4;

	/* Note that "nh == ih", by definition. */
	unsigned char *nh = skb_network_header(skb);
	unsigned int eh_len = nh - data;

	/* The tcp header follows the ip header. */
	struct tcphdr *th = (struct tcphdr *)(nh + ih_len);
	unsigned int th_len = th->doff * 4;

	/* The total number of header bytes. */
	/* NOTE: This may be less than skb_headlen(skb). */
	unsigned int sh_len = eh_len + ih_len + th_len;

	/* The number of payload bytes at "skb->data + sh_len". */
	/* This is non-zero for sendfile() without HIGHDMA. */
	unsigned int b_len = skb_headlen(skb) - sh_len;

	/* The total number of payload bytes. */
	unsigned int d_len = b_len + skb->data_len;

	/* The maximum payload size. */
	unsigned int p_len = sh->gso_size;

	/* The total number of segments. */
	unsigned int num_segs = sh->gso_segs;

	/* The temporary copy of the command. */
	u32 cmd_body[(LEPP_MAX_CMD_SIZE + 3) / 4];
	lepp_tso_cmd_t *cmd = (lepp_tso_cmd_t *)cmd_body;

	/* Analyze the "frags". */
	unsigned int num_frags =
		tile_net_tx_frags(cmd->frags, skb, data + sh_len, b_len);

	/* The size of the command, including frags and header. */
	size_t cmd_size = LEPP_TSO_CMD_SIZE(num_frags, sh_len);

	/* The command header. */
	lepp_tso_cmd_t cmd_init = {
		.tso = true,
		.header_size = sh_len,
		.ip_offset = eh_len,
		.tcp_offset = eh_len + ih_len,
		.payload_size = p_len,
		.num_frags = num_frags,
	};

	unsigned long irqflags;

	lepp_queue_t *eq = priv->eq;

	struct sk_buff *olds[8];
	unsigned int wanted = 8;
	unsigned int i, nolds = 0;

	unsigned int cmd_head, cmd_tail, cmd_next;
	unsigned int comp_tail;


	/* Paranoia. */
	BUG_ON(skb->protocol != htons(ETH_P_IP));
	BUG_ON(ih->protocol != IPPROTO_TCP);
	BUG_ON(skb->ip_summed != CHECKSUM_PARTIAL);
	BUG_ON(num_frags > LEPP_MAX_FRAGS);
	/*--BUG_ON(num_segs != (d_len + (p_len - 1)) / p_len); */
	BUG_ON(num_segs <= 1);


	/* Finish preparing the command. */

	/* Copy the command header. */
	*cmd = cmd_init;

	/* Copy the "header". */
	memcpy(&cmd->frags[num_frags], data, sh_len);


	/* Prefetch and wait, to minimize time spent holding the spinlock. */
	prefetch_L1(&eq->comp_tail);
	prefetch_L1(&eq->cmd_tail);
	mb();


	/* Enqueue the command. */

	spin_lock_irqsave(&priv->eq_lock, irqflags);

	/*
	 * Handle completions if needed to make room.
	 * HACK: Spin until there is sufficient room.
	 */
	if (lepp_num_free_comp_slots(eq) == 0) {
		nolds = tile_net_lepp_grab_comps(eq, olds, wanted, 0);
		if (nolds == 0) {
busy:
			spin_unlock_irqrestore(&priv->eq_lock, irqflags);
			return NETDEV_TX_BUSY;
		}
	}

	cmd_head = eq->cmd_head;
	cmd_tail = eq->cmd_tail;

	/* Prepare to advance, detecting full queue. */
	cmd_next = cmd_tail + cmd_size;
	if (cmd_tail < cmd_head && cmd_next >= cmd_head)
		goto busy;
	if (cmd_next > LEPP_CMD_LIMIT) {
		cmd_next = 0;
		if (cmd_next == cmd_head)
			goto busy;
	}

	/* Copy the command. */
	memcpy(&eq->cmds[cmd_tail], cmd, cmd_size);

	/* Advance. */
	cmd_tail = cmd_next;

	/* Record "skb" for eventual freeing. */
	comp_tail = eq->comp_tail;
	eq->comps[comp_tail] = skb;
	LEPP_QINC(comp_tail);
	eq->comp_tail = comp_tail;

	/* Flush before allowing LEPP to handle the command. */
	/* ISSUE: Is this the optimal location for the flush? */
	__insn_mf();

	eq->cmd_tail = cmd_tail;

	/* NOTE: Using "4" here is more efficient than "0" or "2", */
	/* and, strangely, more efficient than pre-checking the number */
	/* of available completions, and comparing it to 4. */
	if (nolds == 0)
		nolds = tile_net_lepp_grab_comps(eq, olds, wanted, 4);

	spin_unlock_irqrestore(&priv->eq_lock, irqflags);

	/* Handle completions. */
	for (i = 0; i < nolds; i++)
		kfree_skb(olds[i]);

	/* Update stats. */
	stats->tx_packets += num_segs;
	stats->tx_bytes += (num_segs * sh_len) + d_len;

	/* Make sure the egress timer is scheduled. */
	tile_net_schedule_egress_timer(info);

	return NETDEV_TX_OK;
}


/*
 * Transmit a packet (called by the kernel via "hard_start_xmit" hook).
 */
static int tile_net_tx(struct sk_buff *skb, struct net_device *dev)
{
	struct tile_net_priv *priv = netdev_priv(dev);
	int my_cpu = smp_processor_id();
	struct tile_net_cpu *info = priv->cpu[my_cpu];
	struct tile_net_stats_t *stats = &info->stats;

	unsigned long irqflags;

	struct skb_shared_info *sh = skb_shinfo(skb);

	unsigned int len = skb->len;
	unsigned char *data = skb->data;

	unsigned int csum_start = skb_checksum_start_offset(skb);

	lepp_frag_t frags[LEPP_MAX_FRAGS];

	unsigned int num_frags;

	lepp_queue_t *eq = priv->eq;

	struct sk_buff *olds[8];
	unsigned int wanted = 8;
	unsigned int i, nolds = 0;

	unsigned int cmd_size = sizeof(lepp_cmd_t);

	unsigned int cmd_head, cmd_tail, cmd_next;
	unsigned int comp_tail;

	lepp_cmd_t cmds[LEPP_MAX_FRAGS];


	/*
	 * This is paranoia, since we think that if the link doesn't come
	 * up, telling Linux we have no carrier will keep it from trying
	 * to transmit.  If it does, though, we can't execute this routine,
	 * since data structures we depend on aren't set up yet.
	 */
	if (!info->registered)
		return NETDEV_TX_BUSY;


	/* Save the timestamp. */
	dev->trans_start = jiffies;


#ifdef TILE_NET_PARANOIA
#if CHIP_HAS_CBOX_HOME_MAP()
	if (hash_default) {
		HV_PTE pte = *virt_to_pte(current->mm, (unsigned long)data);
		if (hv_pte_get_mode(pte) != HV_PTE_MODE_CACHE_HASH_L3)
			panic("Non-HFH egress buffer! VA=%p Mode=%d PTE=%llx",
			      data, hv_pte_get_mode(pte), hv_pte_val(pte));
	}
#endif
#endif


#ifdef TILE_NET_DUMP_PACKETS
	/* ISSUE: Does not dump the "frags". */
	dump_packet(data, skb_headlen(skb), "tx");
#endif /* TILE_NET_DUMP_PACKETS */


	if (sh->gso_size != 0)
		return tile_net_tx_tso(skb, dev);


	/* Prepare the commands. */

	num_frags = tile_net_tx_frags(frags, skb, data, skb_headlen(skb));

	for (i = 0; i < num_frags; i++) {

		bool final = (i == num_frags - 1);

		lepp_cmd_t cmd = {
			.cpa_lo = frags[i].cpa_lo,
			.cpa_hi = frags[i].cpa_hi,
			.length = frags[i].length,
			.hash_for_home = frags[i].hash_for_home,
			.send_completion = final,
			.end_of_packet = final
		};

		if (i == 0 && skb->ip_summed == CHECKSUM_PARTIAL) {
			cmd.compute_checksum = 1;
			cmd.checksum_data.bits.start_byte = csum_start;
			cmd.checksum_data.bits.count = len - csum_start;
			cmd.checksum_data.bits.destination_byte =
				csum_start + skb->csum_offset;
		}

		cmds[i] = cmd;
	}


	/* Prefetch and wait, to minimize time spent holding the spinlock. */
	prefetch_L1(&eq->comp_tail);
	prefetch_L1(&eq->cmd_tail);
	mb();


	/* Enqueue the commands. */

	spin_lock_irqsave(&priv->eq_lock, irqflags);

	/*
	 * Handle completions if needed to make room.
	 * HACK: Spin until there is sufficient room.
	 */
	if (lepp_num_free_comp_slots(eq) == 0) {
		nolds = tile_net_lepp_grab_comps(eq, olds, wanted, 0);
		if (nolds == 0) {
busy:
			spin_unlock_irqrestore(&priv->eq_lock, irqflags);
			return NETDEV_TX_BUSY;
		}
	}

	cmd_head = eq->cmd_head;
	cmd_tail = eq->cmd_tail;

	/* Copy the commands, or fail. */
	for (i = 0; i < num_frags; i++) {

		/* Prepare to advance, detecting full queue. */
		cmd_next = cmd_tail + cmd_size;
		if (cmd_tail < cmd_head && cmd_next >= cmd_head)
			goto busy;
		if (cmd_next > LEPP_CMD_LIMIT) {
			cmd_next = 0;
			if (cmd_next == cmd_head)
				goto busy;
		}

		/* Copy the command. */
		*(lepp_cmd_t *)&eq->cmds[cmd_tail] = cmds[i];

		/* Advance. */
		cmd_tail = cmd_next;
	}

	/* Record "skb" for eventual freeing. */
	comp_tail = eq->comp_tail;
	eq->comps[comp_tail] = skb;
	LEPP_QINC(comp_tail);
	eq->comp_tail = comp_tail;

	/* Flush before allowing LEPP to handle the command. */
	/* ISSUE: Is this the optimal location for the flush? */
	__insn_mf();

	eq->cmd_tail = cmd_tail;

	/* NOTE: Using "4" here is more efficient than "0" or "2", */
	/* and, strangely, more efficient than pre-checking the number */
	/* of available completions, and comparing it to 4. */
	if (nolds == 0)
		nolds = tile_net_lepp_grab_comps(eq, olds, wanted, 4);

	spin_unlock_irqrestore(&priv->eq_lock, irqflags);

	/* Handle completions. */
	for (i = 0; i < nolds; i++)
		kfree_skb(olds[i]);

	/* HACK: Track "expanded" size for short packets (e.g. 42 < 60). */
	stats->tx_packets++;
	stats->tx_bytes += ((len >= ETH_ZLEN) ? len : ETH_ZLEN);

	/* Make sure the egress timer is scheduled. */
	tile_net_schedule_egress_timer(info);

	return NETDEV_TX_OK;
}


/*
 * Deal with a transmit timeout.
 */
static void tile_net_tx_timeout(struct net_device *dev)
{
	PDEBUG("tile_net_tx_timeout()\n");
	PDEBUG("Transmit timeout at %ld, latency %ld\n", jiffies,
	       jiffies - dev->trans_start);

	/* XXX: ISSUE: This doesn't seem useful for us. */
	netif_wake_queue(dev);
}


/*
 * Ioctl commands.
 */
static int tile_net_ioctl(struct net_device *dev, struct ifreq *rq, int cmd)
{
	return -EOPNOTSUPP;
}


/*
 * Get System Network Statistics.
 *
 * Returns the address of the device statistics structure.
 */
static struct net_device_stats *tile_net_get_stats(struct net_device *dev)
{
	struct tile_net_priv *priv = netdev_priv(dev);
	u32 rx_packets = 0;
	u32 tx_packets = 0;
	u32 rx_bytes = 0;
	u32 tx_bytes = 0;
	int i;

	for_each_online_cpu(i) {
		if (priv->cpu[i]) {
			rx_packets += priv->cpu[i]->stats.rx_packets;
			rx_bytes += priv->cpu[i]->stats.rx_bytes;
			tx_packets += priv->cpu[i]->stats.tx_packets;
			tx_bytes += priv->cpu[i]->stats.tx_bytes;
		}
	}

	priv->stats.rx_packets = rx_packets;
	priv->stats.rx_bytes = rx_bytes;
	priv->stats.tx_packets = tx_packets;
	priv->stats.tx_bytes = tx_bytes;

	return &priv->stats;
}


/*
 * Change the "mtu".
 *
 * The "change_mtu" method is usually not needed.
 * If you need it, it must be like this.
 */
static int tile_net_change_mtu(struct net_device *dev, int new_mtu)
{
	PDEBUG("tile_net_change_mtu()\n");

	/* Check ranges. */
	if ((new_mtu < 68) || (new_mtu > 1500))
		return -EINVAL;

	/* Accept the value. */
	dev->mtu = new_mtu;

	return 0;
}


/*
 * Change the Ethernet Address of the NIC.
 *
 * The hypervisor driver does not support changing MAC address.  However,
 * the IPP does not do anything with the MAC address, so the address which
 * gets used on outgoing packets, and which is accepted on incoming packets,
 * is completely up to the NetIO program or kernel driver which is actually
 * handling them.
 *
 * Returns 0 on success, negative on failure.
 */
static int tile_net_set_mac_address(struct net_device *dev, void *p)
{
	struct sockaddr *addr = p;

	if (!is_valid_ether_addr(addr->sa_data))
		return -EINVAL;

	/* ISSUE: Note that "dev_addr" is now a pointer. */
	memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);

	return 0;
}


/*
 * Obtain the MAC address from the hypervisor.
 * This must be done before opening the device.
 */
static int tile_net_get_mac(struct net_device *dev)
{
	struct tile_net_priv *priv = netdev_priv(dev);

	char hv_dev_name[32];
	int len;

	__netio_getset_offset_t offset = { .word = NETIO_IPP_PARAM_OFF };

	int ret;

	/* For example, "xgbe0". */
	strcpy(hv_dev_name, dev->name);
	len = strlen(hv_dev_name);

	/* For example, "xgbe/0". */
	hv_dev_name[len] = hv_dev_name[len - 1];
	hv_dev_name[len - 1] = '/';
	len++;

	/* For example, "xgbe/0/native_hash". */
	strcpy(hv_dev_name + len, hash_default ? "/native_hash" : "/native");

	/* Get the hypervisor handle for this device. */
	priv->hv_devhdl = hv_dev_open((HV_VirtAddr)hv_dev_name, 0);
	PDEBUG("hv_dev_open(%s) returned %d %p\n",
	       hv_dev_name, priv->hv_devhdl, &priv->hv_devhdl);
	if (priv->hv_devhdl < 0) {
		if (priv->hv_devhdl == HV_ENODEV)
			printk(KERN_DEBUG "Ignoring unconfigured device %s\n",
				 hv_dev_name);
		else
			printk(KERN_DEBUG "hv_dev_open(%s) returned %d\n",
				 hv_dev_name, priv->hv_devhdl);
		return -1;
	}

	/*
	 * Read the hardware address from the hypervisor.
	 * ISSUE: Note that "dev_addr" is now a pointer.
	 */
	offset.bits.class = NETIO_PARAM;
	offset.bits.addr = NETIO_PARAM_MAC;
	ret = hv_dev_pread(priv->hv_devhdl, 0,
			   (HV_VirtAddr)dev->dev_addr, dev->addr_len,
			   offset.word);
	PDEBUG("hv_dev_pread(NETIO_PARAM_MAC) returned %d\n", ret);
	if (ret <= 0) {
		printk(KERN_DEBUG "hv_dev_pread(NETIO_PARAM_MAC) %s failed\n",
		       dev->name);
		/*
		 * Since the device is configured by the hypervisor but we
		 * can't get its MAC address, we are most likely running
		 * the simulator, so let's generate a random MAC address.
		 */
		random_ether_addr(dev->dev_addr);
	}

	return 0;
}

static const struct net_device_ops tile_net_ops = {
	.ndo_open = tile_net_open,
	.ndo_stop = tile_net_stop,
	.ndo_start_xmit = tile_net_tx,
	.ndo_do_ioctl = tile_net_ioctl,
	.ndo_get_stats = tile_net_get_stats,
	.ndo_change_mtu = tile_net_change_mtu,
	.ndo_tx_timeout = tile_net_tx_timeout,
	.ndo_set_mac_address = tile_net_set_mac_address
};


/*
 * The setup function.
 *
 * This uses ether_setup() to assign various fields in dev, including
 * setting IFF_BROADCAST and IFF_MULTICAST, then sets some extra fields.
 */
static void tile_net_setup(struct net_device *dev)
{
	PDEBUG("tile_net_setup()\n");

	ether_setup(dev);

	dev->netdev_ops = &tile_net_ops;

	dev->watchdog_timeo = TILE_NET_TIMEOUT;

	/* We want lockless xmit. */
	dev->features |= NETIF_F_LLTX;

	/* We support hardware tx checksums. */
	dev->features |= NETIF_F_HW_CSUM;

	/* We support scatter/gather. */
	dev->features |= NETIF_F_SG;

	/* We support TSO. */
	dev->features |= NETIF_F_TSO;

#ifdef TILE_NET_GSO
	/* We support GSO. */
	dev->features |= NETIF_F_GSO;
#endif

	if (hash_default)
		dev->features |= NETIF_F_HIGHDMA;

	/* ISSUE: We should support NETIF_F_UFO. */

	dev->tx_queue_len = TILE_NET_TX_QUEUE_LEN;

	dev->mtu = TILE_NET_MTU;
}


/*
 * Allocate the device structure, register the device, and obtain the
 * MAC address from the hypervisor.
 */
static struct net_device *tile_net_dev_init(const char *name)
{
	int ret;
	struct net_device *dev;
	struct tile_net_priv *priv;

	/*
	 * Allocate the device structure.  This allocates "priv", calls
	 * tile_net_setup(), and saves "name".  Normally, "name" is a
	 * template, instantiated by register_netdev(), but not for us.
	 */
	dev = alloc_netdev(sizeof(*priv), name, tile_net_setup);
	if (!dev) {
		pr_err("alloc_netdev(%s) failed\n", name);
		return NULL;
	}

	priv = netdev_priv(dev);

	/* Initialize "priv". */

	memset(priv, 0, sizeof(*priv));

	/* Save "dev" for "tile_net_open_retry()". */
	priv->dev = dev;

	INIT_DELAYED_WORK(&priv->retry_work, tile_net_open_retry);

	spin_lock_init(&priv->eq_lock);

	/* Allocate "eq". */
	priv->eq_pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, EQ_ORDER);
	if (!priv->eq_pages) {
		free_netdev(dev);
		return NULL;
	}
	priv->eq = page_address(priv->eq_pages);

	/* Register the network device. */
	ret = register_netdev(dev);
	if (ret) {
		pr_err("register_netdev %s failed %d\n", dev->name, ret);
		__free_pages(priv->eq_pages, EQ_ORDER);
		free_netdev(dev);
		return NULL;
	}

	/* Get the MAC address. */
	ret = tile_net_get_mac(dev);
	if (ret < 0) {
		unregister_netdev(dev);
		__free_pages(priv->eq_pages, EQ_ORDER);
		free_netdev(dev);
		return NULL;
	}

	return dev;
}


/*
 * Module cleanup.
 *
 * FIXME: If compiled as a module, this module cannot be "unloaded",
 * because the "ingress interrupt handler" is registered permanently.
 */
static void tile_net_cleanup(void)
{
	int i;

	for (i = 0; i < TILE_NET_DEVS; i++) {
		if (tile_net_devs[i]) {
			struct net_device *dev = tile_net_devs[i];
			struct tile_net_priv *priv = netdev_priv(dev);
			unregister_netdev(dev);
			finv_buffer_remote(priv->eq, EQ_SIZE, 0);
			__free_pages(priv->eq_pages, EQ_ORDER);
			free_netdev(dev);
		}
	}
}


/*
 * Module initialization.
 */
static int tile_net_init_module(void)
{
	pr_info("Tilera IPP Net Driver\n");

	tile_net_devs[0] = tile_net_dev_init("xgbe0");
	tile_net_devs[1] = tile_net_dev_init("xgbe1");
	tile_net_devs[2] = tile_net_dev_init("gbe0");
	tile_net_devs[3] = tile_net_dev_init("gbe1");

	return 0;
}


module_init(tile_net_init_module);
module_exit(tile_net_cleanup);


#ifndef MODULE

/*
 * The "network_cpus" boot argument specifies the cpus that are dedicated
 * to handle ingress packets.
 *
 * The parameter should be in the form "network_cpus=m-n[,x-y]", where
 * m, n, x, y are integer numbers that represent the cpus that can be
 * neither a dedicated cpu nor a dataplane cpu.
 */
static int __init network_cpus_setup(char *str)
{
	int rc = cpulist_parse_crop(str, &network_cpus_map);
	if (rc != 0) {
		pr_warning("network_cpus=%s: malformed cpu list\n",
		       str);
	} else {

		/* Remove dedicated cpus. */
		cpumask_and(&network_cpus_map, &network_cpus_map,
			    cpu_possible_mask);


		if (cpumask_empty(&network_cpus_map)) {
			pr_warning("Ignoring network_cpus='%s'.\n",
			       str);
		} else {
			char buf[1024];
			cpulist_scnprintf(buf, sizeof(buf), &network_cpus_map);
			pr_info("Linux network CPUs: %s\n", buf);
			network_cpus_used = true;
		}
	}

	return 0;
}
__setup("network_cpus=", network_cpus_setup);

#endif