summaryrefslogtreecommitdiff
path: root/drivers/net/ethernet/chelsio/cxgb4vf/sge.c
blob: 1bb57d3fbbe8783e32187509081ea3a63991641e (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
/*
 * This file is part of the Chelsio T4 PCI-E SR-IOV Virtual Function Ethernet
 * driver for Linux.
 *
 * Copyright (c) 2009-2010 Chelsio Communications, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/skbuff.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/if_vlan.h>
#include <linux/ip.h>
#include <net/ipv6.h>
#include <net/tcp.h>
#include <linux/dma-mapping.h>
#include <linux/prefetch.h>

#include "t4vf_common.h"
#include "t4vf_defs.h"

#include "../cxgb4/t4_regs.h"
#include "../cxgb4/t4_values.h"
#include "../cxgb4/t4fw_api.h"
#include "../cxgb4/t4_msg.h"

/*
 * Constants ...
 */
enum {
	/*
	 * Egress Queue sizes, producer and consumer indices are all in units
	 * of Egress Context Units bytes.  Note that as far as the hardware is
	 * concerned, the free list is an Egress Queue (the host produces free
	 * buffers which the hardware consumes) and free list entries are
	 * 64-bit PCI DMA addresses.
	 */
	EQ_UNIT = SGE_EQ_IDXSIZE,
	FL_PER_EQ_UNIT = EQ_UNIT / sizeof(__be64),
	TXD_PER_EQ_UNIT = EQ_UNIT / sizeof(__be64),

	/*
	 * Max number of TX descriptors we clean up at a time.  Should be
	 * modest as freeing skbs isn't cheap and it happens while holding
	 * locks.  We just need to free packets faster than they arrive, we
	 * eventually catch up and keep the amortized cost reasonable.
	 */
	MAX_TX_RECLAIM = 16,

	/*
	 * Max number of Rx buffers we replenish at a time.  Again keep this
	 * modest, allocating buffers isn't cheap either.
	 */
	MAX_RX_REFILL = 16,

	/*
	 * Period of the Rx queue check timer.  This timer is infrequent as it
	 * has something to do only when the system experiences severe memory
	 * shortage.
	 */
	RX_QCHECK_PERIOD = (HZ / 2),

	/*
	 * Period of the TX queue check timer and the maximum number of TX
	 * descriptors to be reclaimed by the TX timer.
	 */
	TX_QCHECK_PERIOD = (HZ / 2),
	MAX_TIMER_TX_RECLAIM = 100,

	/*
	 * Suspend an Ethernet TX queue with fewer available descriptors than
	 * this.  We always want to have room for a maximum sized packet:
	 * inline immediate data + MAX_SKB_FRAGS. This is the same as
	 * calc_tx_flits() for a TSO packet with nr_frags == MAX_SKB_FRAGS
	 * (see that function and its helpers for a description of the
	 * calculation).
	 */
	ETHTXQ_MAX_FRAGS = MAX_SKB_FRAGS + 1,
	ETHTXQ_MAX_SGL_LEN = ((3 * (ETHTXQ_MAX_FRAGS-1))/2 +
				   ((ETHTXQ_MAX_FRAGS-1) & 1) +
				   2),
	ETHTXQ_MAX_HDR = (sizeof(struct fw_eth_tx_pkt_vm_wr) +
			  sizeof(struct cpl_tx_pkt_lso_core) +
			  sizeof(struct cpl_tx_pkt_core)) / sizeof(__be64),
	ETHTXQ_MAX_FLITS = ETHTXQ_MAX_SGL_LEN + ETHTXQ_MAX_HDR,

	ETHTXQ_STOP_THRES = 1 + DIV_ROUND_UP(ETHTXQ_MAX_FLITS, TXD_PER_EQ_UNIT),

	/*
	 * Max TX descriptor space we allow for an Ethernet packet to be
	 * inlined into a WR.  This is limited by the maximum value which
	 * we can specify for immediate data in the firmware Ethernet TX
	 * Work Request.
	 */
	MAX_IMM_TX_PKT_LEN = FW_WR_IMMDLEN_M,

	/*
	 * Max size of a WR sent through a control TX queue.
	 */
	MAX_CTRL_WR_LEN = 256,

	/*
	 * Maximum amount of data which we'll ever need to inline into a
	 * TX ring: max(MAX_IMM_TX_PKT_LEN, MAX_CTRL_WR_LEN).
	 */
	MAX_IMM_TX_LEN = (MAX_IMM_TX_PKT_LEN > MAX_CTRL_WR_LEN
			  ? MAX_IMM_TX_PKT_LEN
			  : MAX_CTRL_WR_LEN),

	/*
	 * For incoming packets less than RX_COPY_THRES, we copy the data into
	 * an skb rather than referencing the data.  We allocate enough
	 * in-line room in skb's to accommodate pulling in RX_PULL_LEN bytes
	 * of the data (header).
	 */
	RX_COPY_THRES = 256,
	RX_PULL_LEN = 128,

	/*
	 * Main body length for sk_buffs used for RX Ethernet packets with
	 * fragments.  Should be >= RX_PULL_LEN but possibly bigger to give
	 * pskb_may_pull() some room.
	 */
	RX_SKB_LEN = 512,
};

/*
 * Software state per TX descriptor.
 */
struct tx_sw_desc {
	struct sk_buff *skb;		/* socket buffer of TX data source */
	struct ulptx_sgl *sgl;		/* scatter/gather list in TX Queue */
};

/*
 * Software state per RX Free List descriptor.  We keep track of the allocated
 * FL page, its size, and its PCI DMA address (if the page is mapped).  The FL
 * page size and its PCI DMA mapped state are stored in the low bits of the
 * PCI DMA address as per below.
 */
struct rx_sw_desc {
	struct page *page;		/* Free List page buffer */
	dma_addr_t dma_addr;		/* PCI DMA address (if mapped) */
					/*   and flags (see below) */
};

/*
 * The low bits of rx_sw_desc.dma_addr have special meaning.  Note that the
 * SGE also uses the low 4 bits to determine the size of the buffer.  It uses
 * those bits to index into the SGE_FL_BUFFER_SIZE[index] register array.
 * Since we only use SGE_FL_BUFFER_SIZE0 and SGE_FL_BUFFER_SIZE1, these low 4
 * bits can only contain a 0 or a 1 to indicate which size buffer we're giving
 * to the SGE.  Thus, our software state of "is the buffer mapped for DMA" is
 * maintained in an inverse sense so the hardware never sees that bit high.
 */
enum {
	RX_LARGE_BUF    = 1 << 0,	/* buffer is SGE_FL_BUFFER_SIZE[1] */
	RX_UNMAPPED_BUF = 1 << 1,	/* buffer is not mapped */
};

/**
 *	get_buf_addr - return DMA buffer address of software descriptor
 *	@sdesc: pointer to the software buffer descriptor
 *
 *	Return the DMA buffer address of a software descriptor (stripping out
 *	our low-order flag bits).
 */
static inline dma_addr_t get_buf_addr(const struct rx_sw_desc *sdesc)
{
	return sdesc->dma_addr & ~(dma_addr_t)(RX_LARGE_BUF | RX_UNMAPPED_BUF);
}

/**
 *	is_buf_mapped - is buffer mapped for DMA?
 *	@sdesc: pointer to the software buffer descriptor
 *
 *	Determine whether the buffer associated with a software descriptor in
 *	mapped for DMA or not.
 */
static inline bool is_buf_mapped(const struct rx_sw_desc *sdesc)
{
	return !(sdesc->dma_addr & RX_UNMAPPED_BUF);
}

/**
 *	need_skb_unmap - does the platform need unmapping of sk_buffs?
 *
 *	Returns true if the platform needs sk_buff unmapping.  The compiler
 *	optimizes away unnecessary code if this returns true.
 */
static inline int need_skb_unmap(void)
{
#ifdef CONFIG_NEED_DMA_MAP_STATE
	return 1;
#else
	return 0;
#endif
}

/**
 *	txq_avail - return the number of available slots in a TX queue
 *	@tq: the TX queue
 *
 *	Returns the number of available descriptors in a TX queue.
 */
static inline unsigned int txq_avail(const struct sge_txq *tq)
{
	return tq->size - 1 - tq->in_use;
}

/**
 *	fl_cap - return the capacity of a Free List
 *	@fl: the Free List
 *
 *	Returns the capacity of a Free List.  The capacity is less than the
 *	size because an Egress Queue Index Unit worth of descriptors needs to
 *	be left unpopulated, otherwise the Producer and Consumer indices PIDX
 *	and CIDX will match and the hardware will think the FL is empty.
 */
static inline unsigned int fl_cap(const struct sge_fl *fl)
{
	return fl->size - FL_PER_EQ_UNIT;
}

/**
 *	fl_starving - return whether a Free List is starving.
 *	@adapter: pointer to the adapter
 *	@fl: the Free List
 *
 *	Tests specified Free List to see whether the number of buffers
 *	available to the hardware has falled below our "starvation"
 *	threshold.
 */
static inline bool fl_starving(const struct adapter *adapter,
			       const struct sge_fl *fl)
{
	const struct sge *s = &adapter->sge;

	return fl->avail - fl->pend_cred <= s->fl_starve_thres;
}

/**
 *	map_skb -  map an skb for DMA to the device
 *	@dev: the egress net device
 *	@skb: the packet to map
 *	@addr: a pointer to the base of the DMA mapping array
 *
 *	Map an skb for DMA to the device and return an array of DMA addresses.
 */
static int map_skb(struct device *dev, const struct sk_buff *skb,
		   dma_addr_t *addr)
{
	const skb_frag_t *fp, *end;
	const struct skb_shared_info *si;

	*addr = dma_map_single(dev, skb->data, skb_headlen(skb), DMA_TO_DEVICE);
	if (dma_mapping_error(dev, *addr))
		goto out_err;

	si = skb_shinfo(skb);
	end = &si->frags[si->nr_frags];
	for (fp = si->frags; fp < end; fp++) {
		*++addr = skb_frag_dma_map(dev, fp, 0, skb_frag_size(fp),
					   DMA_TO_DEVICE);
		if (dma_mapping_error(dev, *addr))
			goto unwind;
	}
	return 0;

unwind:
	while (fp-- > si->frags)
		dma_unmap_page(dev, *--addr, skb_frag_size(fp), DMA_TO_DEVICE);
	dma_unmap_single(dev, addr[-1], skb_headlen(skb), DMA_TO_DEVICE);

out_err:
	return -ENOMEM;
}

static void unmap_sgl(struct device *dev, const struct sk_buff *skb,
		      const struct ulptx_sgl *sgl, const struct sge_txq *tq)
{
	const struct ulptx_sge_pair *p;
	unsigned int nfrags = skb_shinfo(skb)->nr_frags;

	if (likely(skb_headlen(skb)))
		dma_unmap_single(dev, be64_to_cpu(sgl->addr0),
				 be32_to_cpu(sgl->len0), DMA_TO_DEVICE);
	else {
		dma_unmap_page(dev, be64_to_cpu(sgl->addr0),
			       be32_to_cpu(sgl->len0), DMA_TO_DEVICE);
		nfrags--;
	}

	/*
	 * the complexity below is because of the possibility of a wrap-around
	 * in the middle of an SGL
	 */
	for (p = sgl->sge; nfrags >= 2; nfrags -= 2) {
		if (likely((u8 *)(p + 1) <= (u8 *)tq->stat)) {
unmap:
			dma_unmap_page(dev, be64_to_cpu(p->addr[0]),
				       be32_to_cpu(p->len[0]), DMA_TO_DEVICE);
			dma_unmap_page(dev, be64_to_cpu(p->addr[1]),
				       be32_to_cpu(p->len[1]), DMA_TO_DEVICE);
			p++;
		} else if ((u8 *)p == (u8 *)tq->stat) {
			p = (const struct ulptx_sge_pair *)tq->desc;
			goto unmap;
		} else if ((u8 *)p + 8 == (u8 *)tq->stat) {
			const __be64 *addr = (const __be64 *)tq->desc;

			dma_unmap_page(dev, be64_to_cpu(addr[0]),
				       be32_to_cpu(p->len[0]), DMA_TO_DEVICE);
			dma_unmap_page(dev, be64_to_cpu(addr[1]),
				       be32_to_cpu(p->len[1]), DMA_TO_DEVICE);
			p = (const struct ulptx_sge_pair *)&addr[2];
		} else {
			const __be64 *addr = (const __be64 *)tq->desc;

			dma_unmap_page(dev, be64_to_cpu(p->addr[0]),
				       be32_to_cpu(p->len[0]), DMA_TO_DEVICE);
			dma_unmap_page(dev, be64_to_cpu(addr[0]),
				       be32_to_cpu(p->len[1]), DMA_TO_DEVICE);
			p = (const struct ulptx_sge_pair *)&addr[1];
		}
	}
	if (nfrags) {
		__be64 addr;

		if ((u8 *)p == (u8 *)tq->stat)
			p = (const struct ulptx_sge_pair *)tq->desc;
		addr = ((u8 *)p + 16 <= (u8 *)tq->stat
			? p->addr[0]
			: *(const __be64 *)tq->desc);
		dma_unmap_page(dev, be64_to_cpu(addr), be32_to_cpu(p->len[0]),
			       DMA_TO_DEVICE);
	}
}

/**
 *	free_tx_desc - reclaims TX descriptors and their buffers
 *	@adapter: the adapter
 *	@tq: the TX queue to reclaim descriptors from
 *	@n: the number of descriptors to reclaim
 *	@unmap: whether the buffers should be unmapped for DMA
 *
 *	Reclaims TX descriptors from an SGE TX queue and frees the associated
 *	TX buffers.  Called with the TX queue lock held.
 */
static void free_tx_desc(struct adapter *adapter, struct sge_txq *tq,
			 unsigned int n, bool unmap)
{
	struct tx_sw_desc *sdesc;
	unsigned int cidx = tq->cidx;
	struct device *dev = adapter->pdev_dev;

	const int need_unmap = need_skb_unmap() && unmap;

	sdesc = &tq->sdesc[cidx];
	while (n--) {
		/*
		 * If we kept a reference to the original TX skb, we need to
		 * unmap it from PCI DMA space (if required) and free it.
		 */
		if (sdesc->skb) {
			if (need_unmap)
				unmap_sgl(dev, sdesc->skb, sdesc->sgl, tq);
			dev_consume_skb_any(sdesc->skb);
			sdesc->skb = NULL;
		}

		sdesc++;
		if (++cidx == tq->size) {
			cidx = 0;
			sdesc = tq->sdesc;
		}
	}
	tq->cidx = cidx;
}

/*
 * Return the number of reclaimable descriptors in a TX queue.
 */
static inline int reclaimable(const struct sge_txq *tq)
{
	int hw_cidx = be16_to_cpu(tq->stat->cidx);
	int reclaimable = hw_cidx - tq->cidx;
	if (reclaimable < 0)
		reclaimable += tq->size;
	return reclaimable;
}

/**
 *	reclaim_completed_tx - reclaims completed TX descriptors
 *	@adapter: the adapter
 *	@tq: the TX queue to reclaim completed descriptors from
 *	@unmap: whether the buffers should be unmapped for DMA
 *
 *	Reclaims TX descriptors that the SGE has indicated it has processed,
 *	and frees the associated buffers if possible.  Called with the TX
 *	queue locked.
 */
static inline void reclaim_completed_tx(struct adapter *adapter,
					struct sge_txq *tq,
					bool unmap)
{
	int avail = reclaimable(tq);

	if (avail) {
		/*
		 * Limit the amount of clean up work we do at a time to keep
		 * the TX lock hold time O(1).
		 */
		if (avail > MAX_TX_RECLAIM)
			avail = MAX_TX_RECLAIM;

		free_tx_desc(adapter, tq, avail, unmap);
		tq->in_use -= avail;
	}
}

/**
 *	get_buf_size - return the size of an RX Free List buffer.
 *	@adapter: pointer to the associated adapter
 *	@sdesc: pointer to the software buffer descriptor
 */
static inline int get_buf_size(const struct adapter *adapter,
			       const struct rx_sw_desc *sdesc)
{
	const struct sge *s = &adapter->sge;

	return (s->fl_pg_order > 0 && (sdesc->dma_addr & RX_LARGE_BUF)
		? (PAGE_SIZE << s->fl_pg_order) : PAGE_SIZE);
}

/**
 *	free_rx_bufs - free RX buffers on an SGE Free List
 *	@adapter: the adapter
 *	@fl: the SGE Free List to free buffers from
 *	@n: how many buffers to free
 *
 *	Release the next @n buffers on an SGE Free List RX queue.   The
 *	buffers must be made inaccessible to hardware before calling this
 *	function.
 */
static void free_rx_bufs(struct adapter *adapter, struct sge_fl *fl, int n)
{
	while (n--) {
		struct rx_sw_desc *sdesc = &fl->sdesc[fl->cidx];

		if (is_buf_mapped(sdesc))
			dma_unmap_page(adapter->pdev_dev, get_buf_addr(sdesc),
				       get_buf_size(adapter, sdesc),
				       PCI_DMA_FROMDEVICE);
		put_page(sdesc->page);
		sdesc->page = NULL;
		if (++fl->cidx == fl->size)
			fl->cidx = 0;
		fl->avail--;
	}
}

/**
 *	unmap_rx_buf - unmap the current RX buffer on an SGE Free List
 *	@adapter: the adapter
 *	@fl: the SGE Free List
 *
 *	Unmap the current buffer on an SGE Free List RX queue.   The
 *	buffer must be made inaccessible to HW before calling this function.
 *
 *	This is similar to @free_rx_bufs above but does not free the buffer.
 *	Do note that the FL still loses any further access to the buffer.
 *	This is used predominantly to "transfer ownership" of an FL buffer
 *	to another entity (typically an skb's fragment list).
 */
static void unmap_rx_buf(struct adapter *adapter, struct sge_fl *fl)
{
	struct rx_sw_desc *sdesc = &fl->sdesc[fl->cidx];

	if (is_buf_mapped(sdesc))
		dma_unmap_page(adapter->pdev_dev, get_buf_addr(sdesc),
			       get_buf_size(adapter, sdesc),
			       PCI_DMA_FROMDEVICE);
	sdesc->page = NULL;
	if (++fl->cidx == fl->size)
		fl->cidx = 0;
	fl->avail--;
}

/**
 *	ring_fl_db - righ doorbell on free list
 *	@adapter: the adapter
 *	@fl: the Free List whose doorbell should be rung ...
 *
 *	Tell the Scatter Gather Engine that there are new free list entries
 *	available.
 */
static inline void ring_fl_db(struct adapter *adapter, struct sge_fl *fl)
{
	u32 val = adapter->params.arch.sge_fl_db;

	/* The SGE keeps track of its Producer and Consumer Indices in terms
	 * of Egress Queue Units so we can only tell it about integral numbers
	 * of multiples of Free List Entries per Egress Queue Units ...
	 */
	if (fl->pend_cred >= FL_PER_EQ_UNIT) {
		if (is_t4(adapter->params.chip))
			val |= PIDX_V(fl->pend_cred / FL_PER_EQ_UNIT);
		else
			val |= PIDX_T5_V(fl->pend_cred / FL_PER_EQ_UNIT);

		/* Make sure all memory writes to the Free List queue are
		 * committed before we tell the hardware about them.
		 */
		wmb();

		/* If we don't have access to the new User Doorbell (T5+), use
		 * the old doorbell mechanism; otherwise use the new BAR2
		 * mechanism.
		 */
		if (unlikely(fl->bar2_addr == NULL)) {
			t4_write_reg(adapter,
				     T4VF_SGE_BASE_ADDR + SGE_VF_KDOORBELL,
				     QID_V(fl->cntxt_id) | val);
		} else {
			writel(val | QID_V(fl->bar2_qid),
			       fl->bar2_addr + SGE_UDB_KDOORBELL);

			/* This Write memory Barrier will force the write to
			 * the User Doorbell area to be flushed.
			 */
			wmb();
		}
		fl->pend_cred %= FL_PER_EQ_UNIT;
	}
}

/**
 *	set_rx_sw_desc - initialize software RX buffer descriptor
 *	@sdesc: pointer to the softwore RX buffer descriptor
 *	@page: pointer to the page data structure backing the RX buffer
 *	@dma_addr: PCI DMA address (possibly with low-bit flags)
 */
static inline void set_rx_sw_desc(struct rx_sw_desc *sdesc, struct page *page,
				  dma_addr_t dma_addr)
{
	sdesc->page = page;
	sdesc->dma_addr = dma_addr;
}

/*
 * Support for poisoning RX buffers ...
 */
#define POISON_BUF_VAL -1

static inline void poison_buf(struct page *page, size_t sz)
{
#if POISON_BUF_VAL >= 0
	memset(page_address(page), POISON_BUF_VAL, sz);
#endif
}

/**
 *	refill_fl - refill an SGE RX buffer ring
 *	@adapter: the adapter
 *	@fl: the Free List ring to refill
 *	@n: the number of new buffers to allocate
 *	@gfp: the gfp flags for the allocations
 *
 *	(Re)populate an SGE free-buffer queue with up to @n new packet buffers,
 *	allocated with the supplied gfp flags.  The caller must assure that
 *	@n does not exceed the queue's capacity -- i.e. (cidx == pidx) _IN
 *	EGRESS QUEUE UNITS_ indicates an empty Free List!  Returns the number
 *	of buffers allocated.  If afterwards the queue is found critically low,
 *	mark it as starving in the bitmap of starving FLs.
 */
static unsigned int refill_fl(struct adapter *adapter, struct sge_fl *fl,
			      int n, gfp_t gfp)
{
	struct sge *s = &adapter->sge;
	struct page *page;
	dma_addr_t dma_addr;
	unsigned int cred = fl->avail;
	__be64 *d = &fl->desc[fl->pidx];
	struct rx_sw_desc *sdesc = &fl->sdesc[fl->pidx];

	/*
	 * Sanity: ensure that the result of adding n Free List buffers
	 * won't result in wrapping the SGE's Producer Index around to
	 * it's Consumer Index thereby indicating an empty Free List ...
	 */
	BUG_ON(fl->avail + n > fl->size - FL_PER_EQ_UNIT);

	gfp |= __GFP_NOWARN;

	/*
	 * If we support large pages, prefer large buffers and fail over to
	 * small pages if we can't allocate large pages to satisfy the refill.
	 * If we don't support large pages, drop directly into the small page
	 * allocation code.
	 */
	if (s->fl_pg_order == 0)
		goto alloc_small_pages;

	while (n) {
		page = __dev_alloc_pages(gfp, s->fl_pg_order);
		if (unlikely(!page)) {
			/*
			 * We've failed inour attempt to allocate a "large
			 * page".  Fail over to the "small page" allocation
			 * below.
			 */
			fl->large_alloc_failed++;
			break;
		}
		poison_buf(page, PAGE_SIZE << s->fl_pg_order);

		dma_addr = dma_map_page(adapter->pdev_dev, page, 0,
					PAGE_SIZE << s->fl_pg_order,
					PCI_DMA_FROMDEVICE);
		if (unlikely(dma_mapping_error(adapter->pdev_dev, dma_addr))) {
			/*
			 * We've run out of DMA mapping space.  Free up the
			 * buffer and return with what we've managed to put
			 * into the free list.  We don't want to fail over to
			 * the small page allocation below in this case
			 * because DMA mapping resources are typically
			 * critical resources once they become scarse.
			 */
			__free_pages(page, s->fl_pg_order);
			goto out;
		}
		dma_addr |= RX_LARGE_BUF;
		*d++ = cpu_to_be64(dma_addr);

		set_rx_sw_desc(sdesc, page, dma_addr);
		sdesc++;

		fl->avail++;
		if (++fl->pidx == fl->size) {
			fl->pidx = 0;
			sdesc = fl->sdesc;
			d = fl->desc;
		}
		n--;
	}

alloc_small_pages:
	while (n--) {
		page = __dev_alloc_page(gfp);
		if (unlikely(!page)) {
			fl->alloc_failed++;
			break;
		}
		poison_buf(page, PAGE_SIZE);

		dma_addr = dma_map_page(adapter->pdev_dev, page, 0, PAGE_SIZE,
				       PCI_DMA_FROMDEVICE);
		if (unlikely(dma_mapping_error(adapter->pdev_dev, dma_addr))) {
			put_page(page);
			break;
		}
		*d++ = cpu_to_be64(dma_addr);

		set_rx_sw_desc(sdesc, page, dma_addr);
		sdesc++;

		fl->avail++;
		if (++fl->pidx == fl->size) {
			fl->pidx = 0;
			sdesc = fl->sdesc;
			d = fl->desc;
		}
	}

out:
	/*
	 * Update our accounting state to incorporate the new Free List
	 * buffers, tell the hardware about them and return the number of
	 * buffers which we were able to allocate.
	 */
	cred = fl->avail - cred;
	fl->pend_cred += cred;
	ring_fl_db(adapter, fl);

	if (unlikely(fl_starving(adapter, fl))) {
		smp_wmb();
		set_bit(fl->cntxt_id, adapter->sge.starving_fl);
	}

	return cred;
}

/*
 * Refill a Free List to its capacity or the Maximum Refill Increment,
 * whichever is smaller ...
 */
static inline void __refill_fl(struct adapter *adapter, struct sge_fl *fl)
{
	refill_fl(adapter, fl,
		  min((unsigned int)MAX_RX_REFILL, fl_cap(fl) - fl->avail),
		  GFP_ATOMIC);
}

/**
 *	alloc_ring - allocate resources for an SGE descriptor ring
 *	@dev: the PCI device's core device
 *	@nelem: the number of descriptors
 *	@hwsize: the size of each hardware descriptor
 *	@swsize: the size of each software descriptor
 *	@busaddrp: the physical PCI bus address of the allocated ring
 *	@swringp: return address pointer for software ring
 *	@stat_size: extra space in hardware ring for status information
 *
 *	Allocates resources for an SGE descriptor ring, such as TX queues,
 *	free buffer lists, response queues, etc.  Each SGE ring requires
 *	space for its hardware descriptors plus, optionally, space for software
 *	state associated with each hardware entry (the metadata).  The function
 *	returns three values: the virtual address for the hardware ring (the
 *	return value of the function), the PCI bus address of the hardware
 *	ring (in *busaddrp), and the address of the software ring (in swringp).
 *	Both the hardware and software rings are returned zeroed out.
 */
static void *alloc_ring(struct device *dev, size_t nelem, size_t hwsize,
			size_t swsize, dma_addr_t *busaddrp, void *swringp,
			size_t stat_size)
{
	/*
	 * Allocate the hardware ring and PCI DMA bus address space for said.
	 */
	size_t hwlen = nelem * hwsize + stat_size;
	void *hwring = dma_alloc_coherent(dev, hwlen, busaddrp, GFP_KERNEL);

	if (!hwring)
		return NULL;

	/*
	 * If the caller wants a software ring, allocate it and return a
	 * pointer to it in *swringp.
	 */
	BUG_ON((swsize != 0) != (swringp != NULL));
	if (swsize) {
		void *swring = kcalloc(nelem, swsize, GFP_KERNEL);

		if (!swring) {
			dma_free_coherent(dev, hwlen, hwring, *busaddrp);
			return NULL;
		}
		*(void **)swringp = swring;
	}

	/*
	 * Zero out the hardware ring and return its address as our function
	 * value.
	 */
	memset(hwring, 0, hwlen);
	return hwring;
}

/**
 *	sgl_len - calculates the size of an SGL of the given capacity
 *	@n: the number of SGL entries
 *
 *	Calculates the number of flits (8-byte units) needed for a Direct
 *	Scatter/Gather List that can hold the given number of entries.
 */
static inline unsigned int sgl_len(unsigned int n)
{
	/*
	 * A Direct Scatter Gather List uses 32-bit lengths and 64-bit PCI DMA
	 * addresses.  The DSGL Work Request starts off with a 32-bit DSGL
	 * ULPTX header, then Length0, then Address0, then, for 1 <= i <= N,
	 * repeated sequences of { Length[i], Length[i+1], Address[i],
	 * Address[i+1] } (this ensures that all addresses are on 64-bit
	 * boundaries).  If N is even, then Length[N+1] should be set to 0 and
	 * Address[N+1] is omitted.
	 *
	 * The following calculation incorporates all of the above.  It's
	 * somewhat hard to follow but, briefly: the "+2" accounts for the
	 * first two flits which include the DSGL header, Length0 and
	 * Address0; the "(3*(n-1))/2" covers the main body of list entries (3
	 * flits for every pair of the remaining N) +1 if (n-1) is odd; and
	 * finally the "+((n-1)&1)" adds the one remaining flit needed if
	 * (n-1) is odd ...
	 */
	n--;
	return (3 * n) / 2 + (n & 1) + 2;
}

/**
 *	flits_to_desc - returns the num of TX descriptors for the given flits
 *	@flits: the number of flits
 *
 *	Returns the number of TX descriptors needed for the supplied number
 *	of flits.
 */
static inline unsigned int flits_to_desc(unsigned int flits)
{
	BUG_ON(flits > SGE_MAX_WR_LEN / sizeof(__be64));
	return DIV_ROUND_UP(flits, TXD_PER_EQ_UNIT);
}

/**
 *	is_eth_imm - can an Ethernet packet be sent as immediate data?
 *	@skb: the packet
 *
 *	Returns whether an Ethernet packet is small enough to fit completely as
 *	immediate data.
 */
static inline int is_eth_imm(const struct sk_buff *skb)
{
	/*
	 * The VF Driver uses the FW_ETH_TX_PKT_VM_WR firmware Work Request
	 * which does not accommodate immediate data.  We could dike out all
	 * of the support code for immediate data but that would tie our hands
	 * too much if we ever want to enhace the firmware.  It would also
	 * create more differences between the PF and VF Drivers.
	 */
	return false;
}

/**
 *	calc_tx_flits - calculate the number of flits for a packet TX WR
 *	@skb: the packet
 *
 *	Returns the number of flits needed for a TX Work Request for the
 *	given Ethernet packet, including the needed WR and CPL headers.
 */
static inline unsigned int calc_tx_flits(const struct sk_buff *skb)
{
	unsigned int flits;

	/*
	 * If the skb is small enough, we can pump it out as a work request
	 * with only immediate data.  In that case we just have to have the
	 * TX Packet header plus the skb data in the Work Request.
	 */
	if (is_eth_imm(skb))
		return DIV_ROUND_UP(skb->len + sizeof(struct cpl_tx_pkt),
				    sizeof(__be64));

	/*
	 * Otherwise, we're going to have to construct a Scatter gather list
	 * of the skb body and fragments.  We also include the flits necessary
	 * for the TX Packet Work Request and CPL.  We always have a firmware
	 * Write Header (incorporated as part of the cpl_tx_pkt_lso and
	 * cpl_tx_pkt structures), followed by either a TX Packet Write CPL
	 * message or, if we're doing a Large Send Offload, an LSO CPL message
	 * with an embedded TX Packet Write CPL message.
	 */
	flits = sgl_len(skb_shinfo(skb)->nr_frags + 1);
	if (skb_shinfo(skb)->gso_size)
		flits += (sizeof(struct fw_eth_tx_pkt_vm_wr) +
			  sizeof(struct cpl_tx_pkt_lso_core) +
			  sizeof(struct cpl_tx_pkt_core)) / sizeof(__be64);
	else
		flits += (sizeof(struct fw_eth_tx_pkt_vm_wr) +
			  sizeof(struct cpl_tx_pkt_core)) / sizeof(__be64);
	return flits;
}

/**
 *	write_sgl - populate a Scatter/Gather List for a packet
 *	@skb: the packet
 *	@tq: the TX queue we are writing into
 *	@sgl: starting location for writing the SGL
 *	@end: points right after the end of the SGL
 *	@start: start offset into skb main-body data to include in the SGL
 *	@addr: the list of DMA bus addresses for the SGL elements
 *
 *	Generates a Scatter/Gather List for the buffers that make up a packet.
 *	The caller must provide adequate space for the SGL that will be written.
 *	The SGL includes all of the packet's page fragments and the data in its
 *	main body except for the first @start bytes.  @pos must be 16-byte
 *	aligned and within a TX descriptor with available space.  @end points
 *	write after the end of the SGL but does not account for any potential
 *	wrap around, i.e., @end > @tq->stat.
 */
static void write_sgl(const struct sk_buff *skb, struct sge_txq *tq,
		      struct ulptx_sgl *sgl, u64 *end, unsigned int start,
		      const dma_addr_t *addr)
{
	unsigned int i, len;
	struct ulptx_sge_pair *to;
	const struct skb_shared_info *si = skb_shinfo(skb);
	unsigned int nfrags = si->nr_frags;
	struct ulptx_sge_pair buf[MAX_SKB_FRAGS / 2 + 1];

	len = skb_headlen(skb) - start;
	if (likely(len)) {
		sgl->len0 = htonl(len);
		sgl->addr0 = cpu_to_be64(addr[0] + start);
		nfrags++;
	} else {
		sgl->len0 = htonl(skb_frag_size(&si->frags[0]));
		sgl->addr0 = cpu_to_be64(addr[1]);
	}

	sgl->cmd_nsge = htonl(ULPTX_CMD_V(ULP_TX_SC_DSGL) |
			      ULPTX_NSGE_V(nfrags));
	if (likely(--nfrags == 0))
		return;
	/*
	 * Most of the complexity below deals with the possibility we hit the
	 * end of the queue in the middle of writing the SGL.  For this case
	 * only we create the SGL in a temporary buffer and then copy it.
	 */
	to = (u8 *)end > (u8 *)tq->stat ? buf : sgl->sge;

	for (i = (nfrags != si->nr_frags); nfrags >= 2; nfrags -= 2, to++) {
		to->len[0] = cpu_to_be32(skb_frag_size(&si->frags[i]));
		to->len[1] = cpu_to_be32(skb_frag_size(&si->frags[++i]));
		to->addr[0] = cpu_to_be64(addr[i]);
		to->addr[1] = cpu_to_be64(addr[++i]);
	}
	if (nfrags) {
		to->len[0] = cpu_to_be32(skb_frag_size(&si->frags[i]));
		to->len[1] = cpu_to_be32(0);
		to->addr[0] = cpu_to_be64(addr[i + 1]);
	}
	if (unlikely((u8 *)end > (u8 *)tq->stat)) {
		unsigned int part0 = (u8 *)tq->stat - (u8 *)sgl->sge, part1;

		if (likely(part0))
			memcpy(sgl->sge, buf, part0);
		part1 = (u8 *)end - (u8 *)tq->stat;
		memcpy(tq->desc, (u8 *)buf + part0, part1);
		end = (void *)tq->desc + part1;
	}
	if ((uintptr_t)end & 8)           /* 0-pad to multiple of 16 */
		*end = 0;
}

/**
 *	check_ring_tx_db - check and potentially ring a TX queue's doorbell
 *	@adapter: the adapter
 *	@tq: the TX queue
 *	@n: number of new descriptors to give to HW
 *
 *	Ring the doorbel for a TX queue.
 */
static inline void ring_tx_db(struct adapter *adapter, struct sge_txq *tq,
			      int n)
{
	/* Make sure that all writes to the TX Descriptors are committed
	 * before we tell the hardware about them.
	 */
	wmb();

	/* If we don't have access to the new User Doorbell (T5+), use the old
	 * doorbell mechanism; otherwise use the new BAR2 mechanism.
	 */
	if (unlikely(tq->bar2_addr == NULL)) {
		u32 val = PIDX_V(n);

		t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_KDOORBELL,
			     QID_V(tq->cntxt_id) | val);
	} else {
		u32 val = PIDX_T5_V(n);

		/* T4 and later chips share the same PIDX field offset within
		 * the doorbell, but T5 and later shrank the field in order to
		 * gain a bit for Doorbell Priority.  The field was absurdly
		 * large in the first place (14 bits) so we just use the T5
		 * and later limits and warn if a Queue ID is too large.
		 */
		WARN_ON(val & DBPRIO_F);

		/* If we're only writing a single Egress Unit and the BAR2
		 * Queue ID is 0, we can use the Write Combining Doorbell
		 * Gather Buffer; otherwise we use the simple doorbell.
		 */
		if (n == 1 && tq->bar2_qid == 0) {
			unsigned int index = (tq->pidx
					      ? (tq->pidx - 1)
					      : (tq->size - 1));
			__be64 *src = (__be64 *)&tq->desc[index];
			__be64 __iomem *dst = (__be64 __iomem *)(tq->bar2_addr +
							 SGE_UDB_WCDOORBELL);
			unsigned int count = EQ_UNIT / sizeof(__be64);

			/* Copy the TX Descriptor in a tight loop in order to
			 * try to get it to the adapter in a single Write
			 * Combined transfer on the PCI-E Bus.  If the Write
			 * Combine fails (say because of an interrupt, etc.)
			 * the hardware will simply take the last write as a
			 * simple doorbell write with a PIDX Increment of 1
			 * and will fetch the TX Descriptor from memory via
			 * DMA.
			 */
			while (count) {
				/* the (__force u64) is because the compiler
				 * doesn't understand the endian swizzling
				 * going on
				 */
				writeq((__force u64)*src, dst);
				src++;
				dst++;
				count--;
			}
		} else
			writel(val | QID_V(tq->bar2_qid),
			       tq->bar2_addr + SGE_UDB_KDOORBELL);

		/* This Write Memory Barrier will force the write to the User
		 * Doorbell area to be flushed.  This is needed to prevent
		 * writes on different CPUs for the same queue from hitting
		 * the adapter out of order.  This is required when some Work
		 * Requests take the Write Combine Gather Buffer path (user
		 * doorbell area offset [SGE_UDB_WCDOORBELL..+63]) and some
		 * take the traditional path where we simply increment the
		 * PIDX (User Doorbell area SGE_UDB_KDOORBELL) and have the
		 * hardware DMA read the actual Work Request.
		 */
		wmb();
	}
}

/**
 *	inline_tx_skb - inline a packet's data into TX descriptors
 *	@skb: the packet
 *	@tq: the TX queue where the packet will be inlined
 *	@pos: starting position in the TX queue to inline the packet
 *
 *	Inline a packet's contents directly into TX descriptors, starting at
 *	the given position within the TX DMA ring.
 *	Most of the complexity of this operation is dealing with wrap arounds
 *	in the middle of the packet we want to inline.
 */
static void inline_tx_skb(const struct sk_buff *skb, const struct sge_txq *tq,
			  void *pos)
{
	u64 *p;
	int left = (void *)tq->stat - pos;

	if (likely(skb->len <= left)) {
		if (likely(!skb->data_len))
			skb_copy_from_linear_data(skb, pos, skb->len);
		else
			skb_copy_bits(skb, 0, pos, skb->len);
		pos += skb->len;
	} else {
		skb_copy_bits(skb, 0, pos, left);
		skb_copy_bits(skb, left, tq->desc, skb->len - left);
		pos = (void *)tq->desc + (skb->len - left);
	}

	/* 0-pad to multiple of 16 */
	p = PTR_ALIGN(pos, 8);
	if ((uintptr_t)p & 8)
		*p = 0;
}

/*
 * Figure out what HW csum a packet wants and return the appropriate control
 * bits.
 */
static u64 hwcsum(enum chip_type chip, const struct sk_buff *skb)
{
	int csum_type;
	const struct iphdr *iph = ip_hdr(skb);

	if (iph->version == 4) {
		if (iph->protocol == IPPROTO_TCP)
			csum_type = TX_CSUM_TCPIP;
		else if (iph->protocol == IPPROTO_UDP)
			csum_type = TX_CSUM_UDPIP;
		else {
nocsum:
			/*
			 * unknown protocol, disable HW csum
			 * and hope a bad packet is detected
			 */
			return TXPKT_L4CSUM_DIS_F;
		}
	} else {
		/*
		 * this doesn't work with extension headers
		 */
		const struct ipv6hdr *ip6h = (const struct ipv6hdr *)iph;

		if (ip6h->nexthdr == IPPROTO_TCP)
			csum_type = TX_CSUM_TCPIP6;
		else if (ip6h->nexthdr == IPPROTO_UDP)
			csum_type = TX_CSUM_UDPIP6;
		else
			goto nocsum;
	}

	if (likely(csum_type >= TX_CSUM_TCPIP)) {
		u64 hdr_len = TXPKT_IPHDR_LEN_V(skb_network_header_len(skb));
		int eth_hdr_len = skb_network_offset(skb) - ETH_HLEN;

		if (chip <= CHELSIO_T5)
			hdr_len |= TXPKT_ETHHDR_LEN_V(eth_hdr_len);
		else
			hdr_len |= T6_TXPKT_ETHHDR_LEN_V(eth_hdr_len);
		return TXPKT_CSUM_TYPE_V(csum_type) | hdr_len;
	} else {
		int start = skb_transport_offset(skb);

		return TXPKT_CSUM_TYPE_V(csum_type) |
			TXPKT_CSUM_START_V(start) |
			TXPKT_CSUM_LOC_V(start + skb->csum_offset);
	}
}

/*
 * Stop an Ethernet TX queue and record that state change.
 */
static void txq_stop(struct sge_eth_txq *txq)
{
	netif_tx_stop_queue(txq->txq);
	txq->q.stops++;
}

/*
 * Advance our software state for a TX queue by adding n in use descriptors.
 */
static inline void txq_advance(struct sge_txq *tq, unsigned int n)
{
	tq->in_use += n;
	tq->pidx += n;
	if (tq->pidx >= tq->size)
		tq->pidx -= tq->size;
}

/**
 *	t4vf_eth_xmit - add a packet to an Ethernet TX queue
 *	@skb: the packet
 *	@dev: the egress net device
 *
 *	Add a packet to an SGE Ethernet TX queue.  Runs with softirqs disabled.
 */
int t4vf_eth_xmit(struct sk_buff *skb, struct net_device *dev)
{
	u32 wr_mid;
	u64 cntrl, *end;
	int qidx, credits, max_pkt_len;
	unsigned int flits, ndesc;
	struct adapter *adapter;
	struct sge_eth_txq *txq;
	const struct port_info *pi;
	struct fw_eth_tx_pkt_vm_wr *wr;
	struct cpl_tx_pkt_core *cpl;
	const struct skb_shared_info *ssi;
	dma_addr_t addr[MAX_SKB_FRAGS + 1];
	const size_t fw_hdr_copy_len = (sizeof(wr->ethmacdst) +
					sizeof(wr->ethmacsrc) +
					sizeof(wr->ethtype) +
					sizeof(wr->vlantci));

	/*
	 * The chip minimum packet length is 10 octets but the firmware
	 * command that we are using requires that we copy the Ethernet header
	 * (including the VLAN tag) into the header so we reject anything
	 * smaller than that ...
	 */
	if (unlikely(skb->len < fw_hdr_copy_len))
		goto out_free;

	/* Discard the packet if the length is greater than mtu */
	max_pkt_len = ETH_HLEN + dev->mtu;
	if (skb_vlan_tag_present(skb))
		max_pkt_len += VLAN_HLEN;
	if (!skb_shinfo(skb)->gso_size && (unlikely(skb->len > max_pkt_len)))
		goto out_free;

	/*
	 * Figure out which TX Queue we're going to use.
	 */
	pi = netdev_priv(dev);
	adapter = pi->adapter;
	qidx = skb_get_queue_mapping(skb);
	BUG_ON(qidx >= pi->nqsets);
	txq = &adapter->sge.ethtxq[pi->first_qset + qidx];

	/*
	 * Take this opportunity to reclaim any TX Descriptors whose DMA
	 * transfers have completed.
	 */
	reclaim_completed_tx(adapter, &txq->q, true);

	/*
	 * Calculate the number of flits and TX Descriptors we're going to
	 * need along with how many TX Descriptors will be left over after
	 * we inject our Work Request.
	 */
	flits = calc_tx_flits(skb);
	ndesc = flits_to_desc(flits);
	credits = txq_avail(&txq->q) - ndesc;

	if (unlikely(credits < 0)) {
		/*
		 * Not enough room for this packet's Work Request.  Stop the
		 * TX Queue and return a "busy" condition.  The queue will get
		 * started later on when the firmware informs us that space
		 * has opened up.
		 */
		txq_stop(txq);
		dev_err(adapter->pdev_dev,
			"%s: TX ring %u full while queue awake!\n",
			dev->name, qidx);
		return NETDEV_TX_BUSY;
	}

	if (!is_eth_imm(skb) &&
	    unlikely(map_skb(adapter->pdev_dev, skb, addr) < 0)) {
		/*
		 * We need to map the skb into PCI DMA space (because it can't
		 * be in-lined directly into the Work Request) and the mapping
		 * operation failed.  Record the error and drop the packet.
		 */
		txq->mapping_err++;
		goto out_free;
	}

	wr_mid = FW_WR_LEN16_V(DIV_ROUND_UP(flits, 2));
	if (unlikely(credits < ETHTXQ_STOP_THRES)) {
		/*
		 * After we're done injecting the Work Request for this
		 * packet, we'll be below our "stop threshold" so stop the TX
		 * Queue now and schedule a request for an SGE Egress Queue
		 * Update message.  The queue will get started later on when
		 * the firmware processes this Work Request and sends us an
		 * Egress Queue Status Update message indicating that space
		 * has opened up.
		 */
		txq_stop(txq);
		wr_mid |= FW_WR_EQUEQ_F | FW_WR_EQUIQ_F;
	}

	/*
	 * Start filling in our Work Request.  Note that we do _not_ handle
	 * the WR Header wrapping around the TX Descriptor Ring.  If our
	 * maximum header size ever exceeds one TX Descriptor, we'll need to
	 * do something else here.
	 */
	BUG_ON(DIV_ROUND_UP(ETHTXQ_MAX_HDR, TXD_PER_EQ_UNIT) > 1);
	wr = (void *)&txq->q.desc[txq->q.pidx];
	wr->equiq_to_len16 = cpu_to_be32(wr_mid);
	wr->r3[0] = cpu_to_be32(0);
	wr->r3[1] = cpu_to_be32(0);
	skb_copy_from_linear_data(skb, (void *)wr->ethmacdst, fw_hdr_copy_len);
	end = (u64 *)wr + flits;

	/*
	 * If this is a Large Send Offload packet we'll put in an LSO CPL
	 * message with an encapsulated TX Packet CPL message.  Otherwise we
	 * just use a TX Packet CPL message.
	 */
	ssi = skb_shinfo(skb);
	if (ssi->gso_size) {
		struct cpl_tx_pkt_lso_core *lso = (void *)(wr + 1);
		bool v6 = (ssi->gso_type & SKB_GSO_TCPV6) != 0;
		int l3hdr_len = skb_network_header_len(skb);
		int eth_xtra_len = skb_network_offset(skb) - ETH_HLEN;

		wr->op_immdlen =
			cpu_to_be32(FW_WR_OP_V(FW_ETH_TX_PKT_VM_WR) |
				    FW_WR_IMMDLEN_V(sizeof(*lso) +
						    sizeof(*cpl)));
		/*
		 * Fill in the LSO CPL message.
		 */
		lso->lso_ctrl =
			cpu_to_be32(LSO_OPCODE_V(CPL_TX_PKT_LSO) |
				    LSO_FIRST_SLICE_F |
				    LSO_LAST_SLICE_F |
				    LSO_IPV6_V(v6) |
				    LSO_ETHHDR_LEN_V(eth_xtra_len / 4) |
				    LSO_IPHDR_LEN_V(l3hdr_len / 4) |
				    LSO_TCPHDR_LEN_V(tcp_hdr(skb)->doff));
		lso->ipid_ofst = cpu_to_be16(0);
		lso->mss = cpu_to_be16(ssi->gso_size);
		lso->seqno_offset = cpu_to_be32(0);
		if (is_t4(adapter->params.chip))
			lso->len = cpu_to_be32(skb->len);
		else
			lso->len = cpu_to_be32(LSO_T5_XFER_SIZE_V(skb->len));

		/*
		 * Set up TX Packet CPL pointer, control word and perform
		 * accounting.
		 */
		cpl = (void *)(lso + 1);

		if (CHELSIO_CHIP_VERSION(adapter->params.chip) <= CHELSIO_T5)
			cntrl = TXPKT_ETHHDR_LEN_V(eth_xtra_len);
		else
			cntrl = T6_TXPKT_ETHHDR_LEN_V(eth_xtra_len);

		cntrl |= TXPKT_CSUM_TYPE_V(v6 ?
					   TX_CSUM_TCPIP6 : TX_CSUM_TCPIP) |
			 TXPKT_IPHDR_LEN_V(l3hdr_len);
		txq->tso++;
		txq->tx_cso += ssi->gso_segs;
	} else {
		int len;

		len = is_eth_imm(skb) ? skb->len + sizeof(*cpl) : sizeof(*cpl);
		wr->op_immdlen =
			cpu_to_be32(FW_WR_OP_V(FW_ETH_TX_PKT_VM_WR) |
				    FW_WR_IMMDLEN_V(len));

		/*
		 * Set up TX Packet CPL pointer, control word and perform
		 * accounting.
		 */
		cpl = (void *)(wr + 1);
		if (skb->ip_summed == CHECKSUM_PARTIAL) {
			cntrl = hwcsum(adapter->params.chip, skb) |
				TXPKT_IPCSUM_DIS_F;
			txq->tx_cso++;
		} else
			cntrl = TXPKT_L4CSUM_DIS_F | TXPKT_IPCSUM_DIS_F;
	}

	/*
	 * If there's a VLAN tag present, add that to the list of things to
	 * do in this Work Request.
	 */
	if (skb_vlan_tag_present(skb)) {
		txq->vlan_ins++;
		cntrl |= TXPKT_VLAN_VLD_F | TXPKT_VLAN_V(skb_vlan_tag_get(skb));
	}

	/*
	 * Fill in the TX Packet CPL message header.
	 */
	cpl->ctrl0 = cpu_to_be32(TXPKT_OPCODE_V(CPL_TX_PKT_XT) |
				 TXPKT_INTF_V(pi->port_id) |
				 TXPKT_PF_V(0));
	cpl->pack = cpu_to_be16(0);
	cpl->len = cpu_to_be16(skb->len);
	cpl->ctrl1 = cpu_to_be64(cntrl);

#ifdef T4_TRACE
	T4_TRACE5(adapter->tb[txq->q.cntxt_id & 7],
		  "eth_xmit: ndesc %u, credits %u, pidx %u, len %u, frags %u",
		  ndesc, credits, txq->q.pidx, skb->len, ssi->nr_frags);
#endif

	/*
	 * Fill in the body of the TX Packet CPL message with either in-lined
	 * data or a Scatter/Gather List.
	 */
	if (is_eth_imm(skb)) {
		/*
		 * In-line the packet's data and free the skb since we don't
		 * need it any longer.
		 */
		inline_tx_skb(skb, &txq->q, cpl + 1);
		dev_consume_skb_any(skb);
	} else {
		/*
		 * Write the skb's Scatter/Gather list into the TX Packet CPL
		 * message and retain a pointer to the skb so we can free it
		 * later when its DMA completes.  (We store the skb pointer
		 * in the Software Descriptor corresponding to the last TX
		 * Descriptor used by the Work Request.)
		 *
		 * The retained skb will be freed when the corresponding TX
		 * Descriptors are reclaimed after their DMAs complete.
		 * However, this could take quite a while since, in general,
		 * the hardware is set up to be lazy about sending DMA
		 * completion notifications to us and we mostly perform TX
		 * reclaims in the transmit routine.
		 *
		 * This is good for performamce but means that we rely on new
		 * TX packets arriving to run the destructors of completed
		 * packets, which open up space in their sockets' send queues.
		 * Sometimes we do not get such new packets causing TX to
		 * stall.  A single UDP transmitter is a good example of this
		 * situation.  We have a clean up timer that periodically
		 * reclaims completed packets but it doesn't run often enough
		 * (nor do we want it to) to prevent lengthy stalls.  A
		 * solution to this problem is to run the destructor early,
		 * after the packet is queued but before it's DMAd.  A con is
		 * that we lie to socket memory accounting, but the amount of
		 * extra memory is reasonable (limited by the number of TX
		 * descriptors), the packets do actually get freed quickly by
		 * new packets almost always, and for protocols like TCP that
		 * wait for acks to really free up the data the extra memory
		 * is even less.  On the positive side we run the destructors
		 * on the sending CPU rather than on a potentially different
		 * completing CPU, usually a good thing.
		 *
		 * Run the destructor before telling the DMA engine about the
		 * packet to make sure it doesn't complete and get freed
		 * prematurely.
		 */
		struct ulptx_sgl *sgl = (struct ulptx_sgl *)(cpl + 1);
		struct sge_txq *tq = &txq->q;
		int last_desc;

		/*
		 * If the Work Request header was an exact multiple of our TX
		 * Descriptor length, then it's possible that the starting SGL
		 * pointer lines up exactly with the end of our TX Descriptor
		 * ring.  If that's the case, wrap around to the beginning
		 * here ...
		 */
		if (unlikely((void *)sgl == (void *)tq->stat)) {
			sgl = (void *)tq->desc;
			end = ((void *)tq->desc + ((void *)end - (void *)tq->stat));
		}

		write_sgl(skb, tq, sgl, end, 0, addr);
		skb_orphan(skb);

		last_desc = tq->pidx + ndesc - 1;
		if (last_desc >= tq->size)
			last_desc -= tq->size;
		tq->sdesc[last_desc].skb = skb;
		tq->sdesc[last_desc].sgl = sgl;
	}

	/*
	 * Advance our internal TX Queue state, tell the hardware about
	 * the new TX descriptors and return success.
	 */
	txq_advance(&txq->q, ndesc);
	netif_trans_update(dev);
	ring_tx_db(adapter, &txq->q, ndesc);
	return NETDEV_TX_OK;

out_free:
	/*
	 * An error of some sort happened.  Free the TX skb and tell the
	 * OS that we've "dealt" with the packet ...
	 */
	dev_kfree_skb_any(skb);
	return NETDEV_TX_OK;
}

/**
 *	copy_frags - copy fragments from gather list into skb_shared_info
 *	@skb: destination skb
 *	@gl: source internal packet gather list
 *	@offset: packet start offset in first page
 *
 *	Copy an internal packet gather list into a Linux skb_shared_info
 *	structure.
 */
static inline void copy_frags(struct sk_buff *skb,
			      const struct pkt_gl *gl,
			      unsigned int offset)
{
	int i;

	/* usually there's just one frag */
	__skb_fill_page_desc(skb, 0, gl->frags[0].page,
			     gl->frags[0].offset + offset,
			     gl->frags[0].size - offset);
	skb_shinfo(skb)->nr_frags = gl->nfrags;
	for (i = 1; i < gl->nfrags; i++)
		__skb_fill_page_desc(skb, i, gl->frags[i].page,
				     gl->frags[i].offset,
				     gl->frags[i].size);

	/* get a reference to the last page, we don't own it */
	get_page(gl->frags[gl->nfrags - 1].page);
}

/**
 *	t4vf_pktgl_to_skb - build an sk_buff from a packet gather list
 *	@gl: the gather list
 *	@skb_len: size of sk_buff main body if it carries fragments
 *	@pull_len: amount of data to move to the sk_buff's main body
 *
 *	Builds an sk_buff from the given packet gather list.  Returns the
 *	sk_buff or %NULL if sk_buff allocation failed.
 */
static struct sk_buff *t4vf_pktgl_to_skb(const struct pkt_gl *gl,
					 unsigned int skb_len,
					 unsigned int pull_len)
{
	struct sk_buff *skb;

	/*
	 * If the ingress packet is small enough, allocate an skb large enough
	 * for all of the data and copy it inline.  Otherwise, allocate an skb
	 * with enough room to pull in the header and reference the rest of
	 * the data via the skb fragment list.
	 *
	 * Below we rely on RX_COPY_THRES being less than the smallest Rx
	 * buff!  size, which is expected since buffers are at least
	 * PAGE_SIZEd.  In this case packets up to RX_COPY_THRES have only one
	 * fragment.
	 */
	if (gl->tot_len <= RX_COPY_THRES) {
		/* small packets have only one fragment */
		skb = alloc_skb(gl->tot_len, GFP_ATOMIC);
		if (unlikely(!skb))
			goto out;
		__skb_put(skb, gl->tot_len);
		skb_copy_to_linear_data(skb, gl->va, gl->tot_len);
	} else {
		skb = alloc_skb(skb_len, GFP_ATOMIC);
		if (unlikely(!skb))
			goto out;
		__skb_put(skb, pull_len);
		skb_copy_to_linear_data(skb, gl->va, pull_len);

		copy_frags(skb, gl, pull_len);
		skb->len = gl->tot_len;
		skb->data_len = skb->len - pull_len;
		skb->truesize += skb->data_len;
	}

out:
	return skb;
}

/**
 *	t4vf_pktgl_free - free a packet gather list
 *	@gl: the gather list
 *
 *	Releases the pages of a packet gather list.  We do not own the last
 *	page on the list and do not free it.
 */
static void t4vf_pktgl_free(const struct pkt_gl *gl)
{
	int frag;

	frag = gl->nfrags - 1;
	while (frag--)
		put_page(gl->frags[frag].page);
}

/**
 *	do_gro - perform Generic Receive Offload ingress packet processing
 *	@rxq: ingress RX Ethernet Queue
 *	@gl: gather list for ingress packet
 *	@pkt: CPL header for last packet fragment
 *
 *	Perform Generic Receive Offload (GRO) ingress packet processing.
 *	We use the standard Linux GRO interfaces for this.
 */
static void do_gro(struct sge_eth_rxq *rxq, const struct pkt_gl *gl,
		   const struct cpl_rx_pkt *pkt)
{
	struct adapter *adapter = rxq->rspq.adapter;
	struct sge *s = &adapter->sge;
	int ret;
	struct sk_buff *skb;

	skb = napi_get_frags(&rxq->rspq.napi);
	if (unlikely(!skb)) {
		t4vf_pktgl_free(gl);
		rxq->stats.rx_drops++;
		return;
	}

	copy_frags(skb, gl, s->pktshift);
	skb->len = gl->tot_len - s->pktshift;
	skb->data_len = skb->len;
	skb->truesize += skb->data_len;
	skb->ip_summed = CHECKSUM_UNNECESSARY;
	skb_record_rx_queue(skb, rxq->rspq.idx);

	if (pkt->vlan_ex) {
		__vlan_hwaccel_put_tag(skb, cpu_to_be16(ETH_P_8021Q),
					be16_to_cpu(pkt->vlan));
		rxq->stats.vlan_ex++;
	}
	ret = napi_gro_frags(&rxq->rspq.napi);

	if (ret == GRO_HELD)
		rxq->stats.lro_pkts++;
	else if (ret == GRO_MERGED || ret == GRO_MERGED_FREE)
		rxq->stats.lro_merged++;
	rxq->stats.pkts++;
	rxq->stats.rx_cso++;
}

/**
 *	t4vf_ethrx_handler - process an ingress ethernet packet
 *	@rspq: the response queue that received the packet
 *	@rsp: the response queue descriptor holding the RX_PKT message
 *	@gl: the gather list of packet fragments
 *
 *	Process an ingress ethernet packet and deliver it to the stack.
 */
int t4vf_ethrx_handler(struct sge_rspq *rspq, const __be64 *rsp,
		       const struct pkt_gl *gl)
{
	struct sk_buff *skb;
	const struct cpl_rx_pkt *pkt = (void *)rsp;
	bool csum_ok = pkt->csum_calc && !pkt->err_vec &&
		       (rspq->netdev->features & NETIF_F_RXCSUM);
	struct sge_eth_rxq *rxq = container_of(rspq, struct sge_eth_rxq, rspq);
	struct adapter *adapter = rspq->adapter;
	struct sge *s = &adapter->sge;

	/*
	 * If this is a good TCP packet and we have Generic Receive Offload
	 * enabled, handle the packet in the GRO path.
	 */
	if ((pkt->l2info & cpu_to_be32(RXF_TCP_F)) &&
	    (rspq->netdev->features & NETIF_F_GRO) && csum_ok &&
	    !pkt->ip_frag) {
		do_gro(rxq, gl, pkt);
		return 0;
	}

	/*
	 * Convert the Packet Gather List into an skb.
	 */
	skb = t4vf_pktgl_to_skb(gl, RX_SKB_LEN, RX_PULL_LEN);
	if (unlikely(!skb)) {
		t4vf_pktgl_free(gl);
		rxq->stats.rx_drops++;
		return 0;
	}
	__skb_pull(skb, s->pktshift);
	skb->protocol = eth_type_trans(skb, rspq->netdev);
	skb_record_rx_queue(skb, rspq->idx);
	rxq->stats.pkts++;

	if (csum_ok && !pkt->err_vec &&
	    (be32_to_cpu(pkt->l2info) & (RXF_UDP_F | RXF_TCP_F))) {
		if (!pkt->ip_frag)
			skb->ip_summed = CHECKSUM_UNNECESSARY;
		else {
			__sum16 c = (__force __sum16)pkt->csum;
			skb->csum = csum_unfold(c);
			skb->ip_summed = CHECKSUM_COMPLETE;
		}
		rxq->stats.rx_cso++;
	} else
		skb_checksum_none_assert(skb);

	if (pkt->vlan_ex) {
		rxq->stats.vlan_ex++;
		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), be16_to_cpu(pkt->vlan));
	}

	netif_receive_skb(skb);

	return 0;
}

/**
 *	is_new_response - check if a response is newly written
 *	@rc: the response control descriptor
 *	@rspq: the response queue
 *
 *	Returns true if a response descriptor contains a yet unprocessed
 *	response.
 */
static inline bool is_new_response(const struct rsp_ctrl *rc,
				   const struct sge_rspq *rspq)
{
	return ((rc->type_gen >> RSPD_GEN_S) & 0x1) == rspq->gen;
}

/**
 *	restore_rx_bufs - put back a packet's RX buffers
 *	@gl: the packet gather list
 *	@fl: the SGE Free List
 *	@nfrags: how many fragments in @si
 *
 *	Called when we find out that the current packet, @si, can't be
 *	processed right away for some reason.  This is a very rare event and
 *	there's no effort to make this suspension/resumption process
 *	particularly efficient.
 *
 *	We implement the suspension by putting all of the RX buffers associated
 *	with the current packet back on the original Free List.  The buffers
 *	have already been unmapped and are left unmapped, we mark them as
 *	unmapped in order to prevent further unmapping attempts.  (Effectively
 *	this function undoes the series of @unmap_rx_buf calls which were done
 *	to create the current packet's gather list.)  This leaves us ready to
 *	restart processing of the packet the next time we start processing the
 *	RX Queue ...
 */
static void restore_rx_bufs(const struct pkt_gl *gl, struct sge_fl *fl,
			    int frags)
{
	struct rx_sw_desc *sdesc;

	while (frags--) {
		if (fl->cidx == 0)
			fl->cidx = fl->size - 1;
		else
			fl->cidx--;
		sdesc = &fl->sdesc[fl->cidx];
		sdesc->page = gl->frags[frags].page;
		sdesc->dma_addr |= RX_UNMAPPED_BUF;
		fl->avail++;
	}
}

/**
 *	rspq_next - advance to the next entry in a response queue
 *	@rspq: the queue
 *
 *	Updates the state of a response queue to advance it to the next entry.
 */
static inline void rspq_next(struct sge_rspq *rspq)
{
	rspq->cur_desc = (void *)rspq->cur_desc + rspq->iqe_len;
	if (unlikely(++rspq->cidx == rspq->size)) {
		rspq->cidx = 0;
		rspq->gen ^= 1;
		rspq->cur_desc = rspq->desc;
	}
}

/**
 *	process_responses - process responses from an SGE response queue
 *	@rspq: the ingress response queue to process
 *	@budget: how many responses can be processed in this round
 *
 *	Process responses from a Scatter Gather Engine response queue up to
 *	the supplied budget.  Responses include received packets as well as
 *	control messages from firmware or hardware.
 *
 *	Additionally choose the interrupt holdoff time for the next interrupt
 *	on this queue.  If the system is under memory shortage use a fairly
 *	long delay to help recovery.
 */
static int process_responses(struct sge_rspq *rspq, int budget)
{
	struct sge_eth_rxq *rxq = container_of(rspq, struct sge_eth_rxq, rspq);
	struct adapter *adapter = rspq->adapter;
	struct sge *s = &adapter->sge;
	int budget_left = budget;

	while (likely(budget_left)) {
		int ret, rsp_type;
		const struct rsp_ctrl *rc;

		rc = (void *)rspq->cur_desc + (rspq->iqe_len - sizeof(*rc));
		if (!is_new_response(rc, rspq))
			break;

		/*
		 * Figure out what kind of response we've received from the
		 * SGE.
		 */
		dma_rmb();
		rsp_type = RSPD_TYPE_G(rc->type_gen);
		if (likely(rsp_type == RSPD_TYPE_FLBUF_X)) {
			struct page_frag *fp;
			struct pkt_gl gl;
			const struct rx_sw_desc *sdesc;
			u32 bufsz, frag;
			u32 len = be32_to_cpu(rc->pldbuflen_qid);

			/*
			 * If we get a "new buffer" message from the SGE we
			 * need to move on to the next Free List buffer.
			 */
			if (len & RSPD_NEWBUF_F) {
				/*
				 * We get one "new buffer" message when we
				 * first start up a queue so we need to ignore
				 * it when our offset into the buffer is 0.
				 */
				if (likely(rspq->offset > 0)) {
					free_rx_bufs(rspq->adapter, &rxq->fl,
						     1);
					rspq->offset = 0;
				}
				len = RSPD_LEN_G(len);
			}
			gl.tot_len = len;

			/*
			 * Gather packet fragments.
			 */
			for (frag = 0, fp = gl.frags; /**/; frag++, fp++) {
				BUG_ON(frag >= MAX_SKB_FRAGS);
				BUG_ON(rxq->fl.avail == 0);
				sdesc = &rxq->fl.sdesc[rxq->fl.cidx];
				bufsz = get_buf_size(adapter, sdesc);
				fp->page = sdesc->page;
				fp->offset = rspq->offset;
				fp->size = min(bufsz, len);
				len -= fp->size;
				if (!len)
					break;
				unmap_rx_buf(rspq->adapter, &rxq->fl);
			}
			gl.nfrags = frag+1;

			/*
			 * Last buffer remains mapped so explicitly make it
			 * coherent for CPU access and start preloading first
			 * cache line ...
			 */
			dma_sync_single_for_cpu(rspq->adapter->pdev_dev,
						get_buf_addr(sdesc),
						fp->size, DMA_FROM_DEVICE);
			gl.va = (page_address(gl.frags[0].page) +
				 gl.frags[0].offset);
			prefetch(gl.va);

			/*
			 * Hand the new ingress packet to the handler for
			 * this Response Queue.
			 */
			ret = rspq->handler(rspq, rspq->cur_desc, &gl);
			if (likely(ret == 0))
				rspq->offset += ALIGN(fp->size, s->fl_align);
			else
				restore_rx_bufs(&gl, &rxq->fl, frag);
		} else if (likely(rsp_type == RSPD_TYPE_CPL_X)) {
			ret = rspq->handler(rspq, rspq->cur_desc, NULL);
		} else {
			WARN_ON(rsp_type > RSPD_TYPE_CPL_X);
			ret = 0;
		}

		if (unlikely(ret)) {
			/*
			 * Couldn't process descriptor, back off for recovery.
			 * We use the SGE's last timer which has the longest
			 * interrupt coalescing value ...
			 */
			const int NOMEM_TIMER_IDX = SGE_NTIMERS-1;
			rspq->next_intr_params =
				QINTR_TIMER_IDX_V(NOMEM_TIMER_IDX);
			break;
		}

		rspq_next(rspq);
		budget_left--;
	}

	/*
	 * If this is a Response Queue with an associated Free List and
	 * at least two Egress Queue units available in the Free List
	 * for new buffer pointers, refill the Free List.
	 */
	if (rspq->offset >= 0 &&
	    fl_cap(&rxq->fl) - rxq->fl.avail >= 2*FL_PER_EQ_UNIT)
		__refill_fl(rspq->adapter, &rxq->fl);
	return budget - budget_left;
}

/**
 *	napi_rx_handler - the NAPI handler for RX processing
 *	@napi: the napi instance
 *	@budget: how many packets we can process in this round
 *
 *	Handler for new data events when using NAPI.  This does not need any
 *	locking or protection from interrupts as data interrupts are off at
 *	this point and other adapter interrupts do not interfere (the latter
 *	in not a concern at all with MSI-X as non-data interrupts then have
 *	a separate handler).
 */
static int napi_rx_handler(struct napi_struct *napi, int budget)
{
	unsigned int intr_params;
	struct sge_rspq *rspq = container_of(napi, struct sge_rspq, napi);
	int work_done = process_responses(rspq, budget);
	u32 val;

	if (likely(work_done < budget)) {
		napi_complete(napi);
		intr_params = rspq->next_intr_params;
		rspq->next_intr_params = rspq->intr_params;
	} else
		intr_params = QINTR_TIMER_IDX_V(SGE_TIMER_UPD_CIDX);

	if (unlikely(work_done == 0))
		rspq->unhandled_irqs++;

	val = CIDXINC_V(work_done) | SEINTARM_V(intr_params);
	/* If we don't have access to the new User GTS (T5+), use the old
	 * doorbell mechanism; otherwise use the new BAR2 mechanism.
	 */
	if (unlikely(!rspq->bar2_addr)) {
		t4_write_reg(rspq->adapter,
			     T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
			     val | INGRESSQID_V((u32)rspq->cntxt_id));
	} else {
		writel(val | INGRESSQID_V(rspq->bar2_qid),
		       rspq->bar2_addr + SGE_UDB_GTS);
		wmb();
	}
	return work_done;
}

/*
 * The MSI-X interrupt handler for an SGE response queue for the NAPI case
 * (i.e., response queue serviced by NAPI polling).
 */
irqreturn_t t4vf_sge_intr_msix(int irq, void *cookie)
{
	struct sge_rspq *rspq = cookie;

	napi_schedule(&rspq->napi);
	return IRQ_HANDLED;
}

/*
 * Process the indirect interrupt entries in the interrupt queue and kick off
 * NAPI for each queue that has generated an entry.
 */
static unsigned int process_intrq(struct adapter *adapter)
{
	struct sge *s = &adapter->sge;
	struct sge_rspq *intrq = &s->intrq;
	unsigned int work_done;
	u32 val;

	spin_lock(&adapter->sge.intrq_lock);
	for (work_done = 0; ; work_done++) {
		const struct rsp_ctrl *rc;
		unsigned int qid, iq_idx;
		struct sge_rspq *rspq;

		/*
		 * Grab the next response from the interrupt queue and bail
		 * out if it's not a new response.
		 */
		rc = (void *)intrq->cur_desc + (intrq->iqe_len - sizeof(*rc));
		if (!is_new_response(rc, intrq))
			break;

		/*
		 * If the response isn't a forwarded interrupt message issue a
		 * error and go on to the next response message.  This should
		 * never happen ...
		 */
		dma_rmb();
		if (unlikely(RSPD_TYPE_G(rc->type_gen) != RSPD_TYPE_INTR_X)) {
			dev_err(adapter->pdev_dev,
				"Unexpected INTRQ response type %d\n",
				RSPD_TYPE_G(rc->type_gen));
			continue;
		}

		/*
		 * Extract the Queue ID from the interrupt message and perform
		 * sanity checking to make sure it really refers to one of our
		 * Ingress Queues which is active and matches the queue's ID.
		 * None of these error conditions should ever happen so we may
		 * want to either make them fatal and/or conditionalized under
		 * DEBUG.
		 */
		qid = RSPD_QID_G(be32_to_cpu(rc->pldbuflen_qid));
		iq_idx = IQ_IDX(s, qid);
		if (unlikely(iq_idx >= MAX_INGQ)) {
			dev_err(adapter->pdev_dev,
				"Ingress QID %d out of range\n", qid);
			continue;
		}
		rspq = s->ingr_map[iq_idx];
		if (unlikely(rspq == NULL)) {
			dev_err(adapter->pdev_dev,
				"Ingress QID %d RSPQ=NULL\n", qid);
			continue;
		}
		if (unlikely(rspq->abs_id != qid)) {
			dev_err(adapter->pdev_dev,
				"Ingress QID %d refers to RSPQ %d\n",
				qid, rspq->abs_id);
			continue;
		}

		/*
		 * Schedule NAPI processing on the indicated Response Queue
		 * and move on to the next entry in the Forwarded Interrupt
		 * Queue.
		 */
		napi_schedule(&rspq->napi);
		rspq_next(intrq);
	}

	val = CIDXINC_V(work_done) | SEINTARM_V(intrq->intr_params);
	/* If we don't have access to the new User GTS (T5+), use the old
	 * doorbell mechanism; otherwise use the new BAR2 mechanism.
	 */
	if (unlikely(!intrq->bar2_addr)) {
		t4_write_reg(adapter, T4VF_SGE_BASE_ADDR + SGE_VF_GTS,
			     val | INGRESSQID_V(intrq->cntxt_id));
	} else {
		writel(val | INGRESSQID_V(intrq->bar2_qid),
		       intrq->bar2_addr + SGE_UDB_GTS);
		wmb();
	}

	spin_unlock(&adapter->sge.intrq_lock);

	return work_done;
}

/*
 * The MSI interrupt handler handles data events from SGE response queues as
 * well as error and other async events as they all use the same MSI vector.
 */
static irqreturn_t t4vf_intr_msi(int irq, void *cookie)
{
	struct adapter *adapter = cookie;

	process_intrq(adapter);
	return IRQ_HANDLED;
}

/**
 *	t4vf_intr_handler - select the top-level interrupt handler
 *	@adapter: the adapter
 *
 *	Selects the top-level interrupt handler based on the type of interrupts
 *	(MSI-X or MSI).
 */
irq_handler_t t4vf_intr_handler(struct adapter *adapter)
{
	BUG_ON((adapter->flags & (USING_MSIX|USING_MSI)) == 0);
	if (adapter->flags & USING_MSIX)
		return t4vf_sge_intr_msix;
	else
		return t4vf_intr_msi;
}

/**
 *	sge_rx_timer_cb - perform periodic maintenance of SGE RX queues
 *	@data: the adapter
 *
 *	Runs periodically from a timer to perform maintenance of SGE RX queues.
 *
 *	a) Replenishes RX queues that have run out due to memory shortage.
 *	Normally new RX buffers are added when existing ones are consumed but
 *	when out of memory a queue can become empty.  We schedule NAPI to do
 *	the actual refill.
 */
static void sge_rx_timer_cb(unsigned long data)
{
	struct adapter *adapter = (struct adapter *)data;
	struct sge *s = &adapter->sge;
	unsigned int i;

	/*
	 * Scan the "Starving Free Lists" flag array looking for any Free
	 * Lists in need of more free buffers.  If we find one and it's not
	 * being actively polled, then bump its "starving" counter and attempt
	 * to refill it.  If we're successful in adding enough buffers to push
	 * the Free List over the starving threshold, then we can clear its
	 * "starving" status.
	 */
	for (i = 0; i < ARRAY_SIZE(s->starving_fl); i++) {
		unsigned long m;

		for (m = s->starving_fl[i]; m; m &= m - 1) {
			unsigned int id = __ffs(m) + i * BITS_PER_LONG;
			struct sge_fl *fl = s->egr_map[id];

			clear_bit(id, s->starving_fl);
			smp_mb__after_atomic();

			/*
			 * Since we are accessing fl without a lock there's a
			 * small probability of a false positive where we
			 * schedule napi but the FL is no longer starving.
			 * No biggie.
			 */
			if (fl_starving(adapter, fl)) {
				struct sge_eth_rxq *rxq;

				rxq = container_of(fl, struct sge_eth_rxq, fl);
				if (napi_reschedule(&rxq->rspq.napi))
					fl->starving++;
				else
					set_bit(id, s->starving_fl);
			}
		}
	}

	/*
	 * Reschedule the next scan for starving Free Lists ...
	 */
	mod_timer(&s->rx_timer, jiffies + RX_QCHECK_PERIOD);
}

/**
 *	sge_tx_timer_cb - perform periodic maintenance of SGE Tx queues
 *	@data: the adapter
 *
 *	Runs periodically from a timer to perform maintenance of SGE TX queues.
 *
 *	b) Reclaims completed Tx packets for the Ethernet queues.  Normally
 *	packets are cleaned up by new Tx packets, this timer cleans up packets
 *	when no new packets are being submitted.  This is essential for pktgen,
 *	at least.
 */
static void sge_tx_timer_cb(unsigned long data)
{
	struct adapter *adapter = (struct adapter *)data;
	struct sge *s = &adapter->sge;
	unsigned int i, budget;

	budget = MAX_TIMER_TX_RECLAIM;
	i = s->ethtxq_rover;
	do {
		struct sge_eth_txq *txq = &s->ethtxq[i];

		if (reclaimable(&txq->q) && __netif_tx_trylock(txq->txq)) {
			int avail = reclaimable(&txq->q);

			if (avail > budget)
				avail = budget;

			free_tx_desc(adapter, &txq->q, avail, true);
			txq->q.in_use -= avail;
			__netif_tx_unlock(txq->txq);

			budget -= avail;
			if (!budget)
				break;
		}

		i++;
		if (i >= s->ethqsets)
			i = 0;
	} while (i != s->ethtxq_rover);
	s->ethtxq_rover = i;

	/*
	 * If we found too many reclaimable packets schedule a timer in the
	 * near future to continue where we left off.  Otherwise the next timer
	 * will be at its normal interval.
	 */
	mod_timer(&s->tx_timer, jiffies + (budget ? TX_QCHECK_PERIOD : 2));
}

/**
 *	bar2_address - return the BAR2 address for an SGE Queue's Registers
 *	@adapter: the adapter
 *	@qid: the SGE Queue ID
 *	@qtype: the SGE Queue Type (Egress or Ingress)
 *	@pbar2_qid: BAR2 Queue ID or 0 for Queue ID inferred SGE Queues
 *
 *	Returns the BAR2 address for the SGE Queue Registers associated with
 *	@qid.  If BAR2 SGE Registers aren't available, returns NULL.  Also
 *	returns the BAR2 Queue ID to be used with writes to the BAR2 SGE
 *	Queue Registers.  If the BAR2 Queue ID is 0, then "Inferred Queue ID"
 *	Registers are supported (e.g. the Write Combining Doorbell Buffer).
 */
static void __iomem *bar2_address(struct adapter *adapter,
				  unsigned int qid,
				  enum t4_bar2_qtype qtype,
				  unsigned int *pbar2_qid)
{
	u64 bar2_qoffset;
	int ret;

	ret = t4vf_bar2_sge_qregs(adapter, qid, qtype,
				  &bar2_qoffset, pbar2_qid);
	if (ret)
		return NULL;

	return adapter->bar2 + bar2_qoffset;
}

/**
 *	t4vf_sge_alloc_rxq - allocate an SGE RX Queue
 *	@adapter: the adapter
 *	@rspq: pointer to to the new rxq's Response Queue to be filled in
 *	@iqasynch: if 0, a normal rspq; if 1, an asynchronous event queue
 *	@dev: the network device associated with the new rspq
 *	@intr_dest: MSI-X vector index (overriden in MSI mode)
 *	@fl: pointer to the new rxq's Free List to be filled in
 *	@hnd: the interrupt handler to invoke for the rspq
 */
int t4vf_sge_alloc_rxq(struct adapter *adapter, struct sge_rspq *rspq,
		       bool iqasynch, struct net_device *dev,
		       int intr_dest,
		       struct sge_fl *fl, rspq_handler_t hnd)
{
	struct sge *s = &adapter->sge;
	struct port_info *pi = netdev_priv(dev);
	struct fw_iq_cmd cmd, rpl;
	int ret, iqandst, flsz = 0;

	/*
	 * If we're using MSI interrupts and we're not initializing the
	 * Forwarded Interrupt Queue itself, then set up this queue for
	 * indirect interrupts to the Forwarded Interrupt Queue.  Obviously
	 * the Forwarded Interrupt Queue must be set up before any other
	 * ingress queue ...
	 */
	if ((adapter->flags & USING_MSI) && rspq != &adapter->sge.intrq) {
		iqandst = SGE_INTRDST_IQ;
		intr_dest = adapter->sge.intrq.abs_id;
	} else
		iqandst = SGE_INTRDST_PCI;

	/*
	 * Allocate the hardware ring for the Response Queue.  The size needs
	 * to be a multiple of 16 which includes the mandatory status entry
	 * (regardless of whether the Status Page capabilities are enabled or
	 * not).
	 */
	rspq->size = roundup(rspq->size, 16);
	rspq->desc = alloc_ring(adapter->pdev_dev, rspq->size, rspq->iqe_len,
				0, &rspq->phys_addr, NULL, 0);
	if (!rspq->desc)
		return -ENOMEM;

	/*
	 * Fill in the Ingress Queue Command.  Note: Ideally this code would
	 * be in t4vf_hw.c but there are so many parameters and dependencies
	 * on our Linux SGE state that we would end up having to pass tons of
	 * parameters.  We'll have to think about how this might be migrated
	 * into OS-independent common code ...
	 */
	memset(&cmd, 0, sizeof(cmd));
	cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP_V(FW_IQ_CMD) |
				    FW_CMD_REQUEST_F |
				    FW_CMD_WRITE_F |
				    FW_CMD_EXEC_F);
	cmd.alloc_to_len16 = cpu_to_be32(FW_IQ_CMD_ALLOC_F |
					 FW_IQ_CMD_IQSTART_F |
					 FW_LEN16(cmd));
	cmd.type_to_iqandstindex =
		cpu_to_be32(FW_IQ_CMD_TYPE_V(FW_IQ_TYPE_FL_INT_CAP) |
			    FW_IQ_CMD_IQASYNCH_V(iqasynch) |
			    FW_IQ_CMD_VIID_V(pi->viid) |
			    FW_IQ_CMD_IQANDST_V(iqandst) |
			    FW_IQ_CMD_IQANUS_V(1) |
			    FW_IQ_CMD_IQANUD_V(SGE_UPDATEDEL_INTR) |
			    FW_IQ_CMD_IQANDSTINDEX_V(intr_dest));
	cmd.iqdroprss_to_iqesize =
		cpu_to_be16(FW_IQ_CMD_IQPCIECH_V(pi->port_id) |
			    FW_IQ_CMD_IQGTSMODE_F |
			    FW_IQ_CMD_IQINTCNTTHRESH_V(rspq->pktcnt_idx) |
			    FW_IQ_CMD_IQESIZE_V(ilog2(rspq->iqe_len) - 4));
	cmd.iqsize = cpu_to_be16(rspq->size);
	cmd.iqaddr = cpu_to_be64(rspq->phys_addr);

	if (fl) {
		enum chip_type chip =
			CHELSIO_CHIP_VERSION(adapter->params.chip);
		/*
		 * Allocate the ring for the hardware free list (with space
		 * for its status page) along with the associated software
		 * descriptor ring.  The free list size needs to be a multiple
		 * of the Egress Queue Unit and at least 2 Egress Units larger
		 * than the SGE's Egress Congrestion Threshold
		 * (fl_starve_thres - 1).
		 */
		if (fl->size < s->fl_starve_thres - 1 + 2 * FL_PER_EQ_UNIT)
			fl->size = s->fl_starve_thres - 1 + 2 * FL_PER_EQ_UNIT;
		fl->size = roundup(fl->size, FL_PER_EQ_UNIT);
		fl->desc = alloc_ring(adapter->pdev_dev, fl->size,
				      sizeof(__be64), sizeof(struct rx_sw_desc),
				      &fl->addr, &fl->sdesc, s->stat_len);
		if (!fl->desc) {
			ret = -ENOMEM;
			goto err;
		}

		/*
		 * Calculate the size of the hardware free list ring plus
		 * Status Page (which the SGE will place after the end of the
		 * free list ring) in Egress Queue Units.
		 */
		flsz = (fl->size / FL_PER_EQ_UNIT +
			s->stat_len / EQ_UNIT);

		/*
		 * Fill in all the relevant firmware Ingress Queue Command
		 * fields for the free list.
		 */
		cmd.iqns_to_fl0congen =
			cpu_to_be32(
				FW_IQ_CMD_FL0HOSTFCMODE_V(SGE_HOSTFCMODE_NONE) |
				FW_IQ_CMD_FL0PACKEN_F |
				FW_IQ_CMD_FL0PADEN_F);

		/* In T6, for egress queue type FL there is internal overhead
		 * of 16B for header going into FLM module.  Hence the maximum
		 * allowed burst size is 448 bytes.  For T4/T5, the hardware
		 * doesn't coalesce fetch requests if more than 64 bytes of
		 * Free List pointers are provided, so we use a 128-byte Fetch
		 * Burst Minimum there (T6 implements coalescing so we can use
		 * the smaller 64-byte value there).
		 */
		cmd.fl0dcaen_to_fl0cidxfthresh =
			cpu_to_be16(
				FW_IQ_CMD_FL0FBMIN_V(chip <= CHELSIO_T5 ?
						     FETCHBURSTMIN_128B_X :
						     FETCHBURSTMIN_64B_X) |
				FW_IQ_CMD_FL0FBMAX_V((chip <= CHELSIO_T5) ?
						     FETCHBURSTMAX_512B_X :
						     FETCHBURSTMAX_256B_X));
		cmd.fl0size = cpu_to_be16(flsz);
		cmd.fl0addr = cpu_to_be64(fl->addr);
	}

	/*
	 * Issue the firmware Ingress Queue Command and extract the results if
	 * it completes successfully.
	 */
	ret = t4vf_wr_mbox(adapter, &cmd, sizeof(cmd), &rpl);
	if (ret)
		goto err;

	netif_napi_add(dev, &rspq->napi, napi_rx_handler, 64);
	rspq->cur_desc = rspq->desc;
	rspq->cidx = 0;
	rspq->gen = 1;
	rspq->next_intr_params = rspq->intr_params;
	rspq->cntxt_id = be16_to_cpu(rpl.iqid);
	rspq->bar2_addr = bar2_address(adapter,
				       rspq->cntxt_id,
				       T4_BAR2_QTYPE_INGRESS,
				       &rspq->bar2_qid);
	rspq->abs_id = be16_to_cpu(rpl.physiqid);
	rspq->size--;			/* subtract status entry */
	rspq->adapter = adapter;
	rspq->netdev = dev;
	rspq->handler = hnd;

	/* set offset to -1 to distinguish ingress queues without FL */
	rspq->offset = fl ? 0 : -1;

	if (fl) {
		fl->cntxt_id = be16_to_cpu(rpl.fl0id);
		fl->avail = 0;
		fl->pend_cred = 0;
		fl->pidx = 0;
		fl->cidx = 0;
		fl->alloc_failed = 0;
		fl->large_alloc_failed = 0;
		fl->starving = 0;

		/* Note, we must initialize the BAR2 Free List User Doorbell
		 * information before refilling the Free List!
		 */
		fl->bar2_addr = bar2_address(adapter,
					     fl->cntxt_id,
					     T4_BAR2_QTYPE_EGRESS,
					     &fl->bar2_qid);

		refill_fl(adapter, fl, fl_cap(fl), GFP_KERNEL);
	}

	return 0;

err:
	/*
	 * An error occurred.  Clean up our partial allocation state and
	 * return the error.
	 */
	if (rspq->desc) {
		dma_free_coherent(adapter->pdev_dev, rspq->size * rspq->iqe_len,
				  rspq->desc, rspq->phys_addr);
		rspq->desc = NULL;
	}
	if (fl && fl->desc) {
		kfree(fl->sdesc);
		fl->sdesc = NULL;
		dma_free_coherent(adapter->pdev_dev, flsz * EQ_UNIT,
				  fl->desc, fl->addr);
		fl->desc = NULL;
	}
	return ret;
}

/**
 *	t4vf_sge_alloc_eth_txq - allocate an SGE Ethernet TX Queue
 *	@adapter: the adapter
 *	@txq: pointer to the new txq to be filled in
 *	@devq: the network TX queue associated with the new txq
 *	@iqid: the relative ingress queue ID to which events relating to
 *		the new txq should be directed
 */
int t4vf_sge_alloc_eth_txq(struct adapter *adapter, struct sge_eth_txq *txq,
			   struct net_device *dev, struct netdev_queue *devq,
			   unsigned int iqid)
{
	struct sge *s = &adapter->sge;
	int ret, nentries;
	struct fw_eq_eth_cmd cmd, rpl;
	struct port_info *pi = netdev_priv(dev);

	/*
	 * Calculate the size of the hardware TX Queue (including the Status
	 * Page on the end of the TX Queue) in units of TX Descriptors.
	 */
	nentries = txq->q.size + s->stat_len / sizeof(struct tx_desc);

	/*
	 * Allocate the hardware ring for the TX ring (with space for its
	 * status page) along with the associated software descriptor ring.
	 */
	txq->q.desc = alloc_ring(adapter->pdev_dev, txq->q.size,
				 sizeof(struct tx_desc),
				 sizeof(struct tx_sw_desc),
				 &txq->q.phys_addr, &txq->q.sdesc, s->stat_len);
	if (!txq->q.desc)
		return -ENOMEM;

	/*
	 * Fill in the Egress Queue Command.  Note: As with the direct use of
	 * the firmware Ingress Queue COmmand above in our RXQ allocation
	 * routine, ideally, this code would be in t4vf_hw.c.  Again, we'll
	 * have to see if there's some reasonable way to parameterize it
	 * into the common code ...
	 */
	memset(&cmd, 0, sizeof(cmd));
	cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP_V(FW_EQ_ETH_CMD) |
				    FW_CMD_REQUEST_F |
				    FW_CMD_WRITE_F |
				    FW_CMD_EXEC_F);
	cmd.alloc_to_len16 = cpu_to_be32(FW_EQ_ETH_CMD_ALLOC_F |
					 FW_EQ_ETH_CMD_EQSTART_F |
					 FW_LEN16(cmd));
	cmd.viid_pkd = cpu_to_be32(FW_EQ_ETH_CMD_AUTOEQUEQE_F |
				   FW_EQ_ETH_CMD_VIID_V(pi->viid));
	cmd.fetchszm_to_iqid =
		cpu_to_be32(FW_EQ_ETH_CMD_HOSTFCMODE_V(SGE_HOSTFCMODE_STPG) |
			    FW_EQ_ETH_CMD_PCIECHN_V(pi->port_id) |
			    FW_EQ_ETH_CMD_IQID_V(iqid));
	cmd.dcaen_to_eqsize =
		cpu_to_be32(FW_EQ_ETH_CMD_FBMIN_V(SGE_FETCHBURSTMIN_64B) |
			    FW_EQ_ETH_CMD_FBMAX_V(SGE_FETCHBURSTMAX_512B) |
			    FW_EQ_ETH_CMD_CIDXFTHRESH_V(
						SGE_CIDXFLUSHTHRESH_32) |
			    FW_EQ_ETH_CMD_EQSIZE_V(nentries));
	cmd.eqaddr = cpu_to_be64(txq->q.phys_addr);

	/*
	 * Issue the firmware Egress Queue Command and extract the results if
	 * it completes successfully.
	 */
	ret = t4vf_wr_mbox(adapter, &cmd, sizeof(cmd), &rpl);
	if (ret) {
		/*
		 * The girmware Ingress Queue Command failed for some reason.
		 * Free up our partial allocation state and return the error.
		 */
		kfree(txq->q.sdesc);
		txq->q.sdesc = NULL;
		dma_free_coherent(adapter->pdev_dev,
				  nentries * sizeof(struct tx_desc),
				  txq->q.desc, txq->q.phys_addr);
		txq->q.desc = NULL;
		return ret;
	}

	txq->q.in_use = 0;
	txq->q.cidx = 0;
	txq->q.pidx = 0;
	txq->q.stat = (void *)&txq->q.desc[txq->q.size];
	txq->q.cntxt_id = FW_EQ_ETH_CMD_EQID_G(be32_to_cpu(rpl.eqid_pkd));
	txq->q.bar2_addr = bar2_address(adapter,
					txq->q.cntxt_id,
					T4_BAR2_QTYPE_EGRESS,
					&txq->q.bar2_qid);
	txq->q.abs_id =
		FW_EQ_ETH_CMD_PHYSEQID_G(be32_to_cpu(rpl.physeqid_pkd));
	txq->txq = devq;
	txq->tso = 0;
	txq->tx_cso = 0;
	txq->vlan_ins = 0;
	txq->q.stops = 0;
	txq->q.restarts = 0;
	txq->mapping_err = 0;
	return 0;
}

/*
 * Free the DMA map resources associated with a TX queue.
 */
static void free_txq(struct adapter *adapter, struct sge_txq *tq)
{
	struct sge *s = &adapter->sge;

	dma_free_coherent(adapter->pdev_dev,
			  tq->size * sizeof(*tq->desc) + s->stat_len,
			  tq->desc, tq->phys_addr);
	tq->cntxt_id = 0;
	tq->sdesc = NULL;
	tq->desc = NULL;
}

/*
 * Free the resources associated with a response queue (possibly including a
 * free list).
 */
static void free_rspq_fl(struct adapter *adapter, struct sge_rspq *rspq,
			 struct sge_fl *fl)
{
	struct sge *s = &adapter->sge;
	unsigned int flid = fl ? fl->cntxt_id : 0xffff;

	t4vf_iq_free(adapter, FW_IQ_TYPE_FL_INT_CAP,
		     rspq->cntxt_id, flid, 0xffff);
	dma_free_coherent(adapter->pdev_dev, (rspq->size + 1) * rspq->iqe_len,
			  rspq->desc, rspq->phys_addr);
	netif_napi_del(&rspq->napi);
	rspq->netdev = NULL;
	rspq->cntxt_id = 0;
	rspq->abs_id = 0;
	rspq->desc = NULL;

	if (fl) {
		free_rx_bufs(adapter, fl, fl->avail);
		dma_free_coherent(adapter->pdev_dev,
				  fl->size * sizeof(*fl->desc) + s->stat_len,
				  fl->desc, fl->addr);
		kfree(fl->sdesc);
		fl->sdesc = NULL;
		fl->cntxt_id = 0;
		fl->desc = NULL;
	}
}

/**
 *	t4vf_free_sge_resources - free SGE resources
 *	@adapter: the adapter
 *
 *	Frees resources used by the SGE queue sets.
 */
void t4vf_free_sge_resources(struct adapter *adapter)
{
	struct sge *s = &adapter->sge;
	struct sge_eth_rxq *rxq = s->ethrxq;
	struct sge_eth_txq *txq = s->ethtxq;
	struct sge_rspq *evtq = &s->fw_evtq;
	struct sge_rspq *intrq = &s->intrq;
	int qs;

	for (qs = 0; qs < adapter->sge.ethqsets; qs++, rxq++, txq++) {
		if (rxq->rspq.desc)
			free_rspq_fl(adapter, &rxq->rspq, &rxq->fl);
		if (txq->q.desc) {
			t4vf_eth_eq_free(adapter, txq->q.cntxt_id);
			free_tx_desc(adapter, &txq->q, txq->q.in_use, true);
			kfree(txq->q.sdesc);
			free_txq(adapter, &txq->q);
		}
	}
	if (evtq->desc)
		free_rspq_fl(adapter, evtq, NULL);
	if (intrq->desc)
		free_rspq_fl(adapter, intrq, NULL);
}

/**
 *	t4vf_sge_start - enable SGE operation
 *	@adapter: the adapter
 *
 *	Start tasklets and timers associated with the DMA engine.
 */
void t4vf_sge_start(struct adapter *adapter)
{
	adapter->sge.ethtxq_rover = 0;
	mod_timer(&adapter->sge.rx_timer, jiffies + RX_QCHECK_PERIOD);
	mod_timer(&adapter->sge.tx_timer, jiffies + TX_QCHECK_PERIOD);
}

/**
 *	t4vf_sge_stop - disable SGE operation
 *	@adapter: the adapter
 *
 *	Stop tasklets and timers associated with the DMA engine.  Note that
 *	this is effective only if measures have been taken to disable any HW
 *	events that may restart them.
 */
void t4vf_sge_stop(struct adapter *adapter)
{
	struct sge *s = &adapter->sge;

	if (s->rx_timer.function)
		del_timer_sync(&s->rx_timer);
	if (s->tx_timer.function)
		del_timer_sync(&s->tx_timer);
}

/**
 *	t4vf_sge_init - initialize SGE
 *	@adapter: the adapter
 *
 *	Performs SGE initialization needed every time after a chip reset.
 *	We do not initialize any of the queue sets here, instead the driver
 *	top-level must request those individually.  We also do not enable DMA
 *	here, that should be done after the queues have been set up.
 */
int t4vf_sge_init(struct adapter *adapter)
{
	struct sge_params *sge_params = &adapter->params.sge;
	u32 fl0 = sge_params->sge_fl_buffer_size[0];
	u32 fl1 = sge_params->sge_fl_buffer_size[1];
	struct sge *s = &adapter->sge;

	/*
	 * Start by vetting the basic SGE parameters which have been set up by
	 * the Physical Function Driver.  Ideally we should be able to deal
	 * with _any_ configuration.  Practice is different ...
	 */
	if (fl0 != PAGE_SIZE || (fl1 != 0 && fl1 <= fl0)) {
		dev_err(adapter->pdev_dev, "bad SGE FL buffer sizes [%d, %d]\n",
			fl0, fl1);
		return -EINVAL;
	}
	if ((sge_params->sge_control & RXPKTCPLMODE_F) !=
	    RXPKTCPLMODE_V(RXPKTCPLMODE_SPLIT_X)) {
		dev_err(adapter->pdev_dev, "bad SGE CPL MODE\n");
		return -EINVAL;
	}

	/*
	 * Now translate the adapter parameters into our internal forms.
	 */
	if (fl1)
		s->fl_pg_order = ilog2(fl1) - PAGE_SHIFT;
	s->stat_len = ((sge_params->sge_control & EGRSTATUSPAGESIZE_F)
			? 128 : 64);
	s->pktshift = PKTSHIFT_G(sge_params->sge_control);
	s->fl_align = t4vf_fl_pkt_align(adapter);

	/* A FL with <= fl_starve_thres buffers is starving and a periodic
	 * timer will attempt to refill it.  This needs to be larger than the
	 * SGE's Egress Congestion Threshold.  If it isn't, then we can get
	 * stuck waiting for new packets while the SGE is waiting for us to
	 * give it more Free List entries.  (Note that the SGE's Egress
	 * Congestion Threshold is in units of 2 Free List pointers.)
	 */
	switch (CHELSIO_CHIP_VERSION(adapter->params.chip)) {
	case CHELSIO_T4:
		s->fl_starve_thres =
		   EGRTHRESHOLD_G(sge_params->sge_congestion_control);
		break;
	case CHELSIO_T5:
		s->fl_starve_thres =
		   EGRTHRESHOLDPACKING_G(sge_params->sge_congestion_control);
		break;
	case CHELSIO_T6:
	default:
		s->fl_starve_thres =
		   T6_EGRTHRESHOLDPACKING_G(sge_params->sge_congestion_control);
		break;
	}
	s->fl_starve_thres = s->fl_starve_thres * 2 + 1;

	/*
	 * Set up tasklet timers.
	 */
	setup_timer(&s->rx_timer, sge_rx_timer_cb, (unsigned long)adapter);
	setup_timer(&s->tx_timer, sge_tx_timer_cb, (unsigned long)adapter);

	/*
	 * Initialize Forwarded Interrupt Queue lock.
	 */
	spin_lock_init(&s->intrq_lock);

	return 0;
}