summaryrefslogtreecommitdiff
path: root/drivers/net/ethernet/chelsio/cxgb4/sge.c
blob: 5e1b314e11af674f8d0f7f1ebf3bed60835dda08 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
/*
 * This file is part of the Chelsio T4 Ethernet driver for Linux.
 *
 * Copyright (c) 2003-2014 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 <linux/dma-mapping.h>
#include <linux/jiffies.h>
#include <linux/prefetch.h>
#include <linux/export.h>
#include <net/ipv6.h>
#include <net/tcp.h>
#include "cxgb4.h"
#include "t4_regs.h"
#include "t4_msg.h"
#include "t4fw_api.h"

/*
 * Rx buffer size.  We use largish buffers if possible but settle for single
 * pages under memory shortage.
 */
#if PAGE_SHIFT >= 16
# define FL_PG_ORDER 0
#else
# define FL_PG_ORDER (16 - PAGE_SHIFT)
#endif

/* RX_PULL_LEN should be <= RX_COPY_THRES */
#define RX_COPY_THRES    256
#define 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.
 */
#define RX_PKT_SKB_LEN   512

/*
 * 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.  Must be >= 2 * TXQ_STOP_THRES.
 */
#define MAX_TX_RECLAIM 16

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

/*
 * 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.
 */
#define RX_QCHECK_PERIOD (HZ / 2)

/*
 * Period of the Tx queue check timer.
 */
#define TX_QCHECK_PERIOD (HZ / 2)

/* SGE Hung Ingress DMA Threshold Warning time (in Hz) and Warning Repeat Rate
 * (in RX_QCHECK_PERIOD multiples).  If we find one of the SGE Ingress DMA
 * State Machines in the same state for this amount of time (in HZ) then we'll
 * issue a warning about a potential hang.  We'll repeat the warning as the
 * SGE Ingress DMA Channel appears to be hung every N RX_QCHECK_PERIODs till
 * the situation clears.  If the situation clears, we'll note that as well.
 */
#define SGE_IDMA_WARN_THRESH (1 * HZ)
#define SGE_IDMA_WARN_REPEAT (20 * RX_QCHECK_PERIOD)

/*
 * Max number of Tx descriptors to be reclaimed by the Tx timer.
 */
#define MAX_TIMER_TX_RECLAIM 100

/*
 * Timer index used when backing off due to memory shortage.
 */
#define NOMEM_TMR_IDX (SGE_NTIMERS - 1)

/*
 * An FL with <= FL_STARVE_THRES buffers is starving and a periodic timer will
 * attempt to refill it.
 */
#define FL_STARVE_THRES 4

/*
 * Suspend an Ethernet Tx queue with fewer available descriptors than this.
 * This is the same as calc_tx_descs() for a TSO packet with
 * nr_frags == MAX_SKB_FRAGS.
 */
#define ETHTXQ_STOP_THRES \
	(1 + DIV_ROUND_UP((3 * MAX_SKB_FRAGS) / 2 + (MAX_SKB_FRAGS & 1), 8))

/*
 * Suspension threshold for non-Ethernet Tx queues.  We require enough room
 * for a full sized WR.
 */
#define TXQ_STOP_THRES (SGE_MAX_WR_LEN / sizeof(struct tx_desc))

/*
 * Max Tx descriptor space we allow for an Ethernet packet to be inlined
 * into a WR.
 */
#define MAX_IMM_TX_PKT_LEN 128

/*
 * Max size of a WR sent through a control Tx queue.
 */
#define MAX_CTRL_WR_LEN SGE_MAX_WR_LEN

struct tx_sw_desc {                /* SW state per Tx descriptor */
	struct sk_buff *skb;
	struct ulptx_sgl *sgl;
};

struct rx_sw_desc {                /* SW state per Rx descriptor */
	struct page *page;
	dma_addr_t dma_addr;
};

/*
 * Rx buffer sizes for "useskbs" Free List buffers (one ingress packet pe skb
 * buffer).  We currently only support two sizes for 1500- and 9000-byte MTUs.
 * We could easily support more but there doesn't seem to be much need for
 * that ...
 */
#define FL_MTU_SMALL 1500
#define FL_MTU_LARGE 9000

static inline unsigned int fl_mtu_bufsize(struct adapter *adapter,
					  unsigned int mtu)
{
	struct sge *s = &adapter->sge;

	return ALIGN(s->pktshift + ETH_HLEN + VLAN_HLEN + mtu, s->fl_align);
}

#define FL_MTU_SMALL_BUFSIZE(adapter) fl_mtu_bufsize(adapter, FL_MTU_SMALL)
#define FL_MTU_LARGE_BUFSIZE(adapter) fl_mtu_bufsize(adapter, FL_MTU_LARGE)

/*
 * Bits 0..3 of rx_sw_desc.dma_addr have special meaning.  The hardware uses
 * these to specify the buffer size as an index into the SGE Free List Buffer
 * Size register array.  We also use bit 4, when the buffer has been unmapped
 * for DMA, but this is of course never sent to the hardware and is only used
 * to prevent double unmappings.  All of the above requires that the Free List
 * Buffers which we allocate have the bottom 5 bits free (0) -- i.e. are
 * 32-byte or or a power of 2 greater in alignment.  Since the SGE's minimal
 * Free List Buffer alignment is 32 bytes, this works out for us ...
 */
enum {
	RX_BUF_FLAGS     = 0x1f,   /* bottom five bits are special */
	RX_BUF_SIZE      = 0x0f,   /* bottom three bits are for buf sizes */
	RX_UNMAPPED_BUF  = 0x10,   /* buffer is not mapped */

	/*
	 * XXX We shouldn't depend on being able to use these indices.
	 * XXX Especially when some other Master PF has initialized the
	 * XXX adapter or we use the Firmware Configuration File.  We
	 * XXX should really search through the Host Buffer Size register
	 * XXX array for the appropriately sized buffer indices.
	 */
	RX_SMALL_PG_BUF  = 0x0,   /* small (PAGE_SIZE) page buffer */
	RX_LARGE_PG_BUF  = 0x1,   /* buffer large (FL_PG_ORDER) page buffer */

	RX_SMALL_MTU_BUF = 0x2,   /* small MTU buffer */
	RX_LARGE_MTU_BUF = 0x3,   /* large MTU buffer */
};

static int timer_pkt_quota[] = {1, 1, 2, 3, 4, 5};
#define MIN_NAPI_WORK  1

static inline dma_addr_t get_buf_addr(const struct rx_sw_desc *d)
{
	return d->dma_addr & ~(dma_addr_t)RX_BUF_FLAGS;
}

static inline bool is_buf_mapped(const struct rx_sw_desc *d)
{
	return !(d->dma_addr & RX_UNMAPPED_BUF);
}

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

/**
 *	fl_cap - return the capacity of a free-buffer list
 *	@fl: the FL
 *
 *	Returns the capacity of a free-buffer list.  The capacity is less than
 *	the size because one descriptor needs to be left unpopulated, otherwise
 *	HW will think the FL is empty.
 */
static inline unsigned int fl_cap(const struct sge_fl *fl)
{
	return fl->size - 8;   /* 1 descriptor = 8 buffers */
}

static inline bool fl_starving(const struct sge_fl *fl)
{
	return fl->avail - fl->pend_cred <= FL_STARVE_THRES;
}

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;
}

#ifdef CONFIG_NEED_DMA_MAP_STATE
static void unmap_skb(struct device *dev, const struct sk_buff *skb,
		      const dma_addr_t *addr)
{
	const skb_frag_t *fp, *end;
	const struct skb_shared_info *si;

	dma_unmap_single(dev, *addr++, skb_headlen(skb), DMA_TO_DEVICE);

	si = skb_shinfo(skb);
	end = &si->frags[si->nr_frags];
	for (fp = si->frags; fp < end; fp++)
		dma_unmap_page(dev, *addr++, skb_frag_size(fp), DMA_TO_DEVICE);
}

/**
 *	deferred_unmap_destructor - unmap a packet when it is freed
 *	@skb: the packet
 *
 *	This is the packet destructor used for Tx packets that need to remain
 *	mapped until they are freed rather than until their Tx descriptors are
 *	freed.
 */
static void deferred_unmap_destructor(struct sk_buff *skb)
{
	unmap_skb(skb->dev->dev.parent, skb, (dma_addr_t *)skb->head);
}
#endif

static void unmap_sgl(struct device *dev, const struct sk_buff *skb,
		      const struct ulptx_sgl *sgl, const struct sge_txq *q)
{
	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), ntohl(sgl->len0),
				 DMA_TO_DEVICE);
	else {
		dma_unmap_page(dev, be64_to_cpu(sgl->addr0), ntohl(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 *)q->stat)) {
unmap:			dma_unmap_page(dev, be64_to_cpu(p->addr[0]),
				       ntohl(p->len[0]), DMA_TO_DEVICE);
			dma_unmap_page(dev, be64_to_cpu(p->addr[1]),
				       ntohl(p->len[1]), DMA_TO_DEVICE);
			p++;
		} else if ((u8 *)p == (u8 *)q->stat) {
			p = (const struct ulptx_sge_pair *)q->desc;
			goto unmap;
		} else if ((u8 *)p + 8 == (u8 *)q->stat) {
			const __be64 *addr = (const __be64 *)q->desc;

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

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

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

/**
 *	free_tx_desc - reclaims Tx descriptors and their buffers
 *	@adapter: the adapter
 *	@q: 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 *adap, struct sge_txq *q,
			 unsigned int n, bool unmap)
{
	struct tx_sw_desc *d;
	unsigned int cidx = q->cidx;
	struct device *dev = adap->pdev_dev;

	d = &q->sdesc[cidx];
	while (n--) {
		if (d->skb) {                       /* an SGL is present */
			if (unmap)
				unmap_sgl(dev, d->skb, d->sgl, q);
			dev_consume_skb_any(d->skb);
			d->skb = NULL;
		}
		++d;
		if (++cidx == q->size) {
			cidx = 0;
			d = q->sdesc;
		}
	}
	q->cidx = cidx;
}

/*
 * Return the number of reclaimable descriptors in a Tx queue.
 */
static inline int reclaimable(const struct sge_txq *q)
{
	int hw_cidx = ntohs(q->stat->cidx);
	hw_cidx -= q->cidx;
	return hw_cidx < 0 ? hw_cidx + q->size : hw_cidx;
}

/**
 *	reclaim_completed_tx - reclaims completed Tx descriptors
 *	@adap: the adapter
 *	@q: 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 *adap, struct sge_txq *q,
					bool unmap)
{
	int avail = reclaimable(q);

	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(adap, q, avail, unmap);
		q->in_use -= avail;
	}
}

static inline int get_buf_size(struct adapter *adapter,
			       const struct rx_sw_desc *d)
{
	struct sge *s = &adapter->sge;
	unsigned int rx_buf_size_idx = d->dma_addr & RX_BUF_SIZE;
	int buf_size;

	switch (rx_buf_size_idx) {
	case RX_SMALL_PG_BUF:
		buf_size = PAGE_SIZE;
		break;

	case RX_LARGE_PG_BUF:
		buf_size = PAGE_SIZE << s->fl_pg_order;
		break;

	case RX_SMALL_MTU_BUF:
		buf_size = FL_MTU_SMALL_BUFSIZE(adapter);
		break;

	case RX_LARGE_MTU_BUF:
		buf_size = FL_MTU_LARGE_BUFSIZE(adapter);
		break;

	default:
		BUG_ON(1);
	}

	return buf_size;
}

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

		if (is_buf_mapped(d))
			dma_unmap_page(adap->pdev_dev, get_buf_addr(d),
				       get_buf_size(adap, d),
				       PCI_DMA_FROMDEVICE);
		put_page(d->page);
		d->page = NULL;
		if (++q->cidx == q->size)
			q->cidx = 0;
		q->avail--;
	}
}

/**
 *	unmap_rx_buf - unmap the current Rx buffer on an SGE free list
 *	@adap: the adapter
 *	@q: the SGE free list
 *
 *	Unmap the current buffer on an SGE free-buffer 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.
 */
static void unmap_rx_buf(struct adapter *adap, struct sge_fl *q)
{
	struct rx_sw_desc *d = &q->sdesc[q->cidx];

	if (is_buf_mapped(d))
		dma_unmap_page(adap->pdev_dev, get_buf_addr(d),
			       get_buf_size(adap, d), PCI_DMA_FROMDEVICE);
	d->page = NULL;
	if (++q->cidx == q->size)
		q->cidx = 0;
	q->avail--;
}

static inline void ring_fl_db(struct adapter *adap, struct sge_fl *q)
{
	u32 val;
	if (q->pend_cred >= 8) {
		val = PIDX(q->pend_cred / 8);
		if (!is_t4(adap->params.chip))
			val |= DBTYPE(1);
		val |= DBPRIO(1);
		wmb();

		/* If we're on T4, use the old doorbell mechanism; otherwise
		 * use the new BAR2 mechanism.
		 */
		if (is_t4(adap->params.chip)) {
			t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
				     val | QID(q->cntxt_id));
		} else {
			writel(val,  adap->bar2 + q->udb + SGE_UDB_KDOORBELL);

			/* This Write memory Barrier will force the write to
			 * the User Doorbell area to be flushed.
			 */
			wmb();
		}
		q->pend_cred &= 7;
	}
}

static inline void set_rx_sw_desc(struct rx_sw_desc *sd, struct page *pg,
				  dma_addr_t mapping)
{
	sd->page = pg;
	sd->dma_addr = mapping;      /* includes size low bits */
}

/**
 *	refill_fl - refill an SGE Rx buffer ring
 *	@adap: the adapter
 *	@q: the 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.  If afterwards the queue is
 *	found critically low mark it as starving in the bitmap of starving FLs.
 *
 *	Returns the number of buffers allocated.
 */
static unsigned int refill_fl(struct adapter *adap, struct sge_fl *q, int n,
			      gfp_t gfp)
{
	struct sge *s = &adap->sge;
	struct page *pg;
	dma_addr_t mapping;
	unsigned int cred = q->avail;
	__be64 *d = &q->desc[q->pidx];
	struct rx_sw_desc *sd = &q->sdesc[q->pidx];

	gfp |= __GFP_NOWARN | __GFP_COLD;

	if (s->fl_pg_order == 0)
		goto alloc_small_pages;

	/*
	 * Prefer large buffers
	 */
	while (n) {
		pg = alloc_pages(gfp | __GFP_COMP, s->fl_pg_order);
		if (unlikely(!pg)) {
			q->large_alloc_failed++;
			break;       /* fall back to single pages */
		}

		mapping = dma_map_page(adap->pdev_dev, pg, 0,
				       PAGE_SIZE << s->fl_pg_order,
				       PCI_DMA_FROMDEVICE);
		if (unlikely(dma_mapping_error(adap->pdev_dev, mapping))) {
			__free_pages(pg, s->fl_pg_order);
			goto out;   /* do not try small pages for this error */
		}
		mapping |= RX_LARGE_PG_BUF;
		*d++ = cpu_to_be64(mapping);

		set_rx_sw_desc(sd, pg, mapping);
		sd++;

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

alloc_small_pages:
	while (n--) {
		pg = __skb_alloc_page(gfp, NULL);
		if (unlikely(!pg)) {
			q->alloc_failed++;
			break;
		}

		mapping = dma_map_page(adap->pdev_dev, pg, 0, PAGE_SIZE,
				       PCI_DMA_FROMDEVICE);
		if (unlikely(dma_mapping_error(adap->pdev_dev, mapping))) {
			put_page(pg);
			goto out;
		}
		*d++ = cpu_to_be64(mapping);

		set_rx_sw_desc(sd, pg, mapping);
		sd++;

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

out:	cred = q->avail - cred;
	q->pend_cred += cred;
	ring_fl_db(adap, q);

	if (unlikely(fl_starving(q))) {
		smp_wmb();
		set_bit(q->cntxt_id - adap->sge.egr_start,
			adap->sge.starving_fl);
	}

	return cred;
}

static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl)
{
	refill_fl(adap, fl, min(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
 *	@elem_size: the size of each descriptor
 *	@sw_size: the size of the SW state associated with each ring element
 *	@phys: the physical address of the allocated ring
 *	@metadata: address of the array holding the SW state for the ring
 *	@stat_size: extra space in HW ring for status information
 *	@node: preferred node for memory allocations
 *
 *	Allocates resources for an SGE descriptor ring, such as Tx queues,
 *	free buffer lists, or response queues.  Each SGE ring requires
 *	space for its HW descriptors plus, optionally, space for the SW state
 *	associated with each HW entry (the metadata).  The function returns
 *	three values: the virtual address for the HW ring (the return value
 *	of the function), the bus address of the HW ring, and the address
 *	of the SW ring.
 */
static void *alloc_ring(struct device *dev, size_t nelem, size_t elem_size,
			size_t sw_size, dma_addr_t *phys, void *metadata,
			size_t stat_size, int node)
{
	size_t len = nelem * elem_size + stat_size;
	void *s = NULL;
	void *p = dma_alloc_coherent(dev, len, phys, GFP_KERNEL);

	if (!p)
		return NULL;
	if (sw_size) {
		s = kzalloc_node(nelem * sw_size, GFP_KERNEL, node);

		if (!s) {
			dma_free_coherent(dev, len, p, *phys);
			return NULL;
		}
	}
	if (metadata)
		*(void **)metadata = s;
	memset(p, 0, len);
	return p;
}

/**
 *	sgl_len - calculates the size of an SGL of the given capacity
 *	@n: the number of SGL entries
 *
 *	Calculates the number of flits needed for a scatter/gather list that
 *	can hold the given number of entries.
 */
static inline unsigned int sgl_len(unsigned int n)
{
	n--;
	return (3 * n) / 2 + (n & 1) + 2;
}

/**
 *	flits_to_desc - returns the num of Tx descriptors for the given flits
 *	@n: 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 n)
{
	BUG_ON(n > SGE_MAX_WR_LEN / 8);
	return DIV_ROUND_UP(n, 8);
}

/**
 *	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 as
 *	immediate data. Return value corresponds to headroom required.
 */
static inline int is_eth_imm(const struct sk_buff *skb)
{
	int hdrlen = skb_shinfo(skb)->gso_size ?
			sizeof(struct cpl_tx_pkt_lso_core) : 0;

	hdrlen += sizeof(struct cpl_tx_pkt);
	if (skb->len <= MAX_IMM_TX_PKT_LEN - hdrlen)
		return hdrlen;
	return 0;
}

/**
 *	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 WR 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;
	int hdrlen = is_eth_imm(skb);

	if (hdrlen)
		return DIV_ROUND_UP(skb->len + hdrlen, sizeof(__be64));

	flits = sgl_len(skb_shinfo(skb)->nr_frags + 1) + 4;
	if (skb_shinfo(skb)->gso_size)
		flits += 2;
	return flits;
}

/**
 *	calc_tx_descs - calculate the number of Tx descriptors for a packet
 *	@skb: the packet
 *
 *	Returns the number of Tx descriptors needed for the given Ethernet
 *	packet, including the needed WR and CPL headers.
 */
static inline unsigned int calc_tx_descs(const struct sk_buff *skb)
{
	return flits_to_desc(calc_tx_flits(skb));
}

/**
 *	write_sgl - populate a scatter/gather list for a packet
 *	@skb: the packet
 *	@q: 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 bus addresses for the SGL elements
 *
 *	Generates a 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.  @sgl must be 16-byte
 *	aligned and within a Tx descriptor with available space.  @end points
 *	right after the end of the SGL but does not account for any potential
 *	wrap around, i.e., @end > @sgl.
 */
static void write_sgl(const struct sk_buff *skb, struct sge_txq *q,
		      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(ULP_TX_SC_DSGL) | ULPTX_NSGE(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 *)q->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 *)q->stat)) {
		unsigned int part0 = (u8 *)q->stat - (u8 *)sgl->sge, part1;

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

/* This function copies a tx_desc struct to memory mapped BAR2 space(user space
 * writes). For coalesced WR SGE, fetches data from the FIFO instead of from
 * Host.
 */
static void cxgb_pio_copy(u64 __iomem *dst, struct tx_desc *desc)
{
	int count = sizeof(*desc) / sizeof(u64);
	u64 *src = (u64 *)desc;

	while (count) {
		writeq(*src, dst);
		src++;
		dst++;
		count--;
	}
}

/**
 *	ring_tx_db - check and potentially ring a Tx queue's doorbell
 *	@adap: the adapter
 *	@q: 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 *adap, struct sge_txq *q, int n)
{
	wmb();            /* write descriptors before telling HW */

	if (is_t4(adap->params.chip)) {
		u32 val = PIDX(n);
		unsigned long flags;

		/* For T4 we need to participate in the Doorbell Recovery
		 * mechanism.
		 */
		spin_lock_irqsave(&q->db_lock, flags);
		if (!q->db_disabled)
			t4_write_reg(adap, MYPF_REG(SGE_PF_KDOORBELL),
				     QID(q->cntxt_id) | val);
		else
			q->db_pidx_inc += n;
		q->db_pidx = q->pidx;
		spin_unlock_irqrestore(&q->db_lock, flags);
	} else {
		u32 val = PIDX_T5(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(1));

		/* For T5 and later we use the Write-Combine mapped BAR2 User
		 * Doorbell mechanism.  If we're only writing a single TX
		 * Descriptor and TX Write Combining hasn't been disabled, we
		 * can use the Write Combining Gather Buffer; otherwise we use
		 * the simple doorbell.
		 */
		if (n == 1) {
			int index = (q->pidx
				     ? (q->pidx - 1)
				     : (q->size - 1));

			cxgb_pio_copy(adap->bar2 + q->udb + SGE_UDB_WCDOORBELL,
				      q->desc + index);
		} else {
			writel(val,  adap->bar2 + q->udb + 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
 *	@q: the Tx queue where the packet will be inlined
 *	@pos: starting position in the Tx queue where 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 *q,
			  void *pos)
{
	u64 *p;
	int left = (void *)q->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, q->desc, skb->len - left);
		pos = (void *)q->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(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;
		}
	} 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))
		return TXPKT_CSUM_TYPE(csum_type) |
			TXPKT_IPHDR_LEN(skb_network_header_len(skb)) |
			TXPKT_ETHHDR_LEN(skb_network_offset(skb) - ETH_HLEN);
	else {
		int start = skb_transport_offset(skb);

		return TXPKT_CSUM_TYPE(csum_type) | TXPKT_CSUM_START(start) |
			TXPKT_CSUM_LOC(start + skb->csum_offset);
	}
}

static void eth_txq_stop(struct sge_eth_txq *q)
{
	netif_tx_stop_queue(q->txq);
	q->q.stops++;
}

static inline void txq_advance(struct sge_txq *q, unsigned int n)
{
	q->in_use += n;
	q->pidx += n;
	if (q->pidx >= q->size)
		q->pidx -= q->size;
}

/**
 *	t4_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.
 */
netdev_tx_t t4_eth_xmit(struct sk_buff *skb, struct net_device *dev)
{
	int len;
	u32 wr_mid;
	u64 cntrl, *end;
	int qidx, credits;
	unsigned int flits, ndesc;
	struct adapter *adap;
	struct sge_eth_txq *q;
	const struct port_info *pi;
	struct fw_eth_tx_pkt_wr *wr;
	struct cpl_tx_pkt_core *cpl;
	const struct skb_shared_info *ssi;
	dma_addr_t addr[MAX_SKB_FRAGS + 1];
	bool immediate = false;

	/*
	 * The chip min packet length is 10 octets but play safe and reject
	 * anything shorter than an Ethernet header.
	 */
	if (unlikely(skb->len < ETH_HLEN)) {
out_free:	dev_kfree_skb_any(skb);
		return NETDEV_TX_OK;
	}

	pi = netdev_priv(dev);
	adap = pi->adapter;
	qidx = skb_get_queue_mapping(skb);
	q = &adap->sge.ethtxq[qidx + pi->first_qset];

	reclaim_completed_tx(adap, &q->q, true);

	flits = calc_tx_flits(skb);
	ndesc = flits_to_desc(flits);
	credits = txq_avail(&q->q) - ndesc;

	if (unlikely(credits < 0)) {
		eth_txq_stop(q);
		dev_err(adap->pdev_dev,
			"%s: Tx ring %u full while queue awake!\n",
			dev->name, qidx);
		return NETDEV_TX_BUSY;
	}

	if (is_eth_imm(skb))
		immediate = true;

	if (!immediate &&
	    unlikely(map_skb(adap->pdev_dev, skb, addr) < 0)) {
		q->mapping_err++;
		goto out_free;
	}

	wr_mid = FW_WR_LEN16(DIV_ROUND_UP(flits, 2));
	if (unlikely(credits < ETHTXQ_STOP_THRES)) {
		eth_txq_stop(q);
		wr_mid |= FW_WR_EQUEQ | FW_WR_EQUIQ;
	}

	wr = (void *)&q->q.desc[q->q.pidx];
	wr->equiq_to_len16 = htonl(wr_mid);
	wr->r3 = cpu_to_be64(0);
	end = (u64 *)wr + flits;

	len = immediate ? skb->len : 0;
	ssi = skb_shinfo(skb);
	if (ssi->gso_size) {
		struct cpl_tx_pkt_lso *lso = (void *)wr;
		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;

		len += sizeof(*lso);
		wr->op_immdlen = htonl(FW_WR_OP(FW_ETH_TX_PKT_WR) |
				       FW_WR_IMMDLEN(len));
		lso->c.lso_ctrl = htonl(LSO_OPCODE(CPL_TX_PKT_LSO) |
					LSO_FIRST_SLICE | LSO_LAST_SLICE |
					LSO_IPV6(v6) |
					LSO_ETHHDR_LEN(eth_xtra_len / 4) |
					LSO_IPHDR_LEN(l3hdr_len / 4) |
					LSO_TCPHDR_LEN(tcp_hdr(skb)->doff));
		lso->c.ipid_ofst = htons(0);
		lso->c.mss = htons(ssi->gso_size);
		lso->c.seqno_offset = htonl(0);
		if (is_t4(adap->params.chip))
			lso->c.len = htonl(skb->len);
		else
			lso->c.len = htonl(LSO_T5_XFER_SIZE(skb->len));
		cpl = (void *)(lso + 1);
		cntrl = TXPKT_CSUM_TYPE(v6 ? TX_CSUM_TCPIP6 : TX_CSUM_TCPIP) |
			TXPKT_IPHDR_LEN(l3hdr_len) |
			TXPKT_ETHHDR_LEN(eth_xtra_len);
		q->tso++;
		q->tx_cso += ssi->gso_segs;
	} else {
		len += sizeof(*cpl);
		wr->op_immdlen = htonl(FW_WR_OP(FW_ETH_TX_PKT_WR) |
				       FW_WR_IMMDLEN(len));
		cpl = (void *)(wr + 1);
		if (skb->ip_summed == CHECKSUM_PARTIAL) {
			cntrl = hwcsum(skb) | TXPKT_IPCSUM_DIS;
			q->tx_cso++;
		} else
			cntrl = TXPKT_L4CSUM_DIS | TXPKT_IPCSUM_DIS;
	}

	if (vlan_tx_tag_present(skb)) {
		q->vlan_ins++;
		cntrl |= TXPKT_VLAN_VLD | TXPKT_VLAN(vlan_tx_tag_get(skb));
	}

	cpl->ctrl0 = htonl(TXPKT_OPCODE(CPL_TX_PKT_XT) |
			   TXPKT_INTF(pi->tx_chan) | TXPKT_PF(adap->fn));
	cpl->pack = htons(0);
	cpl->len = htons(skb->len);
	cpl->ctrl1 = cpu_to_be64(cntrl);

	if (immediate) {
		inline_tx_skb(skb, &q->q, cpl + 1);
		dev_consume_skb_any(skb);
	} else {
		int last_desc;

		write_sgl(skb, &q->q, (struct ulptx_sgl *)(cpl + 1), end, 0,
			  addr);
		skb_orphan(skb);

		last_desc = q->q.pidx + ndesc - 1;
		if (last_desc >= q->q.size)
			last_desc -= q->q.size;
		q->q.sdesc[last_desc].skb = skb;
		q->q.sdesc[last_desc].sgl = (struct ulptx_sgl *)(cpl + 1);
	}

	txq_advance(&q->q, ndesc);

	ring_tx_db(adap, &q->q, ndesc);
	return NETDEV_TX_OK;
}

/**
 *	reclaim_completed_tx_imm - reclaim completed control-queue Tx descs
 *	@q: the SGE control Tx queue
 *
 *	This is a variant of reclaim_completed_tx() that is used for Tx queues
 *	that send only immediate data (presently just the control queues) and
 *	thus do not have any sk_buffs to release.
 */
static inline void reclaim_completed_tx_imm(struct sge_txq *q)
{
	int hw_cidx = ntohs(q->stat->cidx);
	int reclaim = hw_cidx - q->cidx;

	if (reclaim < 0)
		reclaim += q->size;

	q->in_use -= reclaim;
	q->cidx = hw_cidx;
}

/**
 *	is_imm - check whether a packet can be sent as immediate data
 *	@skb: the packet
 *
 *	Returns true if a packet can be sent as a WR with immediate data.
 */
static inline int is_imm(const struct sk_buff *skb)
{
	return skb->len <= MAX_CTRL_WR_LEN;
}

/**
 *	ctrlq_check_stop - check if a control queue is full and should stop
 *	@q: the queue
 *	@wr: most recent WR written to the queue
 *
 *	Check if a control queue has become full and should be stopped.
 *	We clean up control queue descriptors very lazily, only when we are out.
 *	If the queue is still full after reclaiming any completed descriptors
 *	we suspend it and have the last WR wake it up.
 */
static void ctrlq_check_stop(struct sge_ctrl_txq *q, struct fw_wr_hdr *wr)
{
	reclaim_completed_tx_imm(&q->q);
	if (unlikely(txq_avail(&q->q) < TXQ_STOP_THRES)) {
		wr->lo |= htonl(FW_WR_EQUEQ | FW_WR_EQUIQ);
		q->q.stops++;
		q->full = 1;
	}
}

/**
 *	ctrl_xmit - send a packet through an SGE control Tx queue
 *	@q: the control queue
 *	@skb: the packet
 *
 *	Send a packet through an SGE control Tx queue.  Packets sent through
 *	a control queue must fit entirely as immediate data.
 */
static int ctrl_xmit(struct sge_ctrl_txq *q, struct sk_buff *skb)
{
	unsigned int ndesc;
	struct fw_wr_hdr *wr;

	if (unlikely(!is_imm(skb))) {
		WARN_ON(1);
		dev_kfree_skb(skb);
		return NET_XMIT_DROP;
	}

	ndesc = DIV_ROUND_UP(skb->len, sizeof(struct tx_desc));
	spin_lock(&q->sendq.lock);

	if (unlikely(q->full)) {
		skb->priority = ndesc;                  /* save for restart */
		__skb_queue_tail(&q->sendq, skb);
		spin_unlock(&q->sendq.lock);
		return NET_XMIT_CN;
	}

	wr = (struct fw_wr_hdr *)&q->q.desc[q->q.pidx];
	inline_tx_skb(skb, &q->q, wr);

	txq_advance(&q->q, ndesc);
	if (unlikely(txq_avail(&q->q) < TXQ_STOP_THRES))
		ctrlq_check_stop(q, wr);

	ring_tx_db(q->adap, &q->q, ndesc);
	spin_unlock(&q->sendq.lock);

	kfree_skb(skb);
	return NET_XMIT_SUCCESS;
}

/**
 *	restart_ctrlq - restart a suspended control queue
 *	@data: the control queue to restart
 *
 *	Resumes transmission on a suspended Tx control queue.
 */
static void restart_ctrlq(unsigned long data)
{
	struct sk_buff *skb;
	unsigned int written = 0;
	struct sge_ctrl_txq *q = (struct sge_ctrl_txq *)data;

	spin_lock(&q->sendq.lock);
	reclaim_completed_tx_imm(&q->q);
	BUG_ON(txq_avail(&q->q) < TXQ_STOP_THRES);  /* q should be empty */

	while ((skb = __skb_dequeue(&q->sendq)) != NULL) {
		struct fw_wr_hdr *wr;
		unsigned int ndesc = skb->priority;     /* previously saved */

		/*
		 * Write descriptors and free skbs outside the lock to limit
		 * wait times.  q->full is still set so new skbs will be queued.
		 */
		spin_unlock(&q->sendq.lock);

		wr = (struct fw_wr_hdr *)&q->q.desc[q->q.pidx];
		inline_tx_skb(skb, &q->q, wr);
		kfree_skb(skb);

		written += ndesc;
		txq_advance(&q->q, ndesc);
		if (unlikely(txq_avail(&q->q) < TXQ_STOP_THRES)) {
			unsigned long old = q->q.stops;

			ctrlq_check_stop(q, wr);
			if (q->q.stops != old) {          /* suspended anew */
				spin_lock(&q->sendq.lock);
				goto ringdb;
			}
		}
		if (written > 16) {
			ring_tx_db(q->adap, &q->q, written);
			written = 0;
		}
		spin_lock(&q->sendq.lock);
	}
	q->full = 0;
ringdb: if (written)
		ring_tx_db(q->adap, &q->q, written);
	spin_unlock(&q->sendq.lock);
}

/**
 *	t4_mgmt_tx - send a management message
 *	@adap: the adapter
 *	@skb: the packet containing the management message
 *
 *	Send a management message through control queue 0.
 */
int t4_mgmt_tx(struct adapter *adap, struct sk_buff *skb)
{
	int ret;

	local_bh_disable();
	ret = ctrl_xmit(&adap->sge.ctrlq[0], skb);
	local_bh_enable();
	return ret;
}

/**
 *	is_ofld_imm - check whether a packet can be sent as immediate data
 *	@skb: the packet
 *
 *	Returns true if a packet can be sent as an offload WR with immediate
 *	data.  We currently use the same limit as for Ethernet packets.
 */
static inline int is_ofld_imm(const struct sk_buff *skb)
{
	return skb->len <= MAX_IMM_TX_PKT_LEN;
}

/**
 *	calc_tx_flits_ofld - calculate # of flits for an offload packet
 *	@skb: the packet
 *
 *	Returns the number of flits needed for the given offload packet.
 *	These packets are already fully constructed and no additional headers
 *	will be added.
 */
static inline unsigned int calc_tx_flits_ofld(const struct sk_buff *skb)
{
	unsigned int flits, cnt;

	if (is_ofld_imm(skb))
		return DIV_ROUND_UP(skb->len, 8);

	flits = skb_transport_offset(skb) / 8U;   /* headers */
	cnt = skb_shinfo(skb)->nr_frags;
	if (skb_tail_pointer(skb) != skb_transport_header(skb))
		cnt++;
	return flits + sgl_len(cnt);
}

/**
 *	txq_stop_maperr - stop a Tx queue due to I/O MMU exhaustion
 *	@adap: the adapter
 *	@q: the queue to stop
 *
 *	Mark a Tx queue stopped due to I/O MMU exhaustion and resulting
 *	inability to map packets.  A periodic timer attempts to restart
 *	queues so marked.
 */
static void txq_stop_maperr(struct sge_ofld_txq *q)
{
	q->mapping_err++;
	q->q.stops++;
	set_bit(q->q.cntxt_id - q->adap->sge.egr_start,
		q->adap->sge.txq_maperr);
}

/**
 *	ofldtxq_stop - stop an offload Tx queue that has become full
 *	@q: the queue to stop
 *	@skb: the packet causing the queue to become full
 *
 *	Stops an offload Tx queue that has become full and modifies the packet
 *	being written to request a wakeup.
 */
static void ofldtxq_stop(struct sge_ofld_txq *q, struct sk_buff *skb)
{
	struct fw_wr_hdr *wr = (struct fw_wr_hdr *)skb->data;

	wr->lo |= htonl(FW_WR_EQUEQ | FW_WR_EQUIQ);
	q->q.stops++;
	q->full = 1;
}

/**
 *	service_ofldq - restart a suspended offload queue
 *	@q: the offload queue
 *
 *	Services an offload Tx queue by moving packets from its packet queue
 *	to the HW Tx ring.  The function starts and ends with the queue locked.
 */
static void service_ofldq(struct sge_ofld_txq *q)
{
	u64 *pos;
	int credits;
	struct sk_buff *skb;
	unsigned int written = 0;
	unsigned int flits, ndesc;

	while ((skb = skb_peek(&q->sendq)) != NULL && !q->full) {
		/*
		 * We drop the lock but leave skb on sendq, thus retaining
		 * exclusive access to the state of the queue.
		 */
		spin_unlock(&q->sendq.lock);

		reclaim_completed_tx(q->adap, &q->q, false);

		flits = skb->priority;                /* previously saved */
		ndesc = flits_to_desc(flits);
		credits = txq_avail(&q->q) - ndesc;
		BUG_ON(credits < 0);
		if (unlikely(credits < TXQ_STOP_THRES))
			ofldtxq_stop(q, skb);

		pos = (u64 *)&q->q.desc[q->q.pidx];
		if (is_ofld_imm(skb))
			inline_tx_skb(skb, &q->q, pos);
		else if (map_skb(q->adap->pdev_dev, skb,
				 (dma_addr_t *)skb->head)) {
			txq_stop_maperr(q);
			spin_lock(&q->sendq.lock);
			break;
		} else {
			int last_desc, hdr_len = skb_transport_offset(skb);

			memcpy(pos, skb->data, hdr_len);
			write_sgl(skb, &q->q, (void *)pos + hdr_len,
				  pos + flits, hdr_len,
				  (dma_addr_t *)skb->head);
#ifdef CONFIG_NEED_DMA_MAP_STATE
			skb->dev = q->adap->port[0];
			skb->destructor = deferred_unmap_destructor;
#endif
			last_desc = q->q.pidx + ndesc - 1;
			if (last_desc >= q->q.size)
				last_desc -= q->q.size;
			q->q.sdesc[last_desc].skb = skb;
		}

		txq_advance(&q->q, ndesc);
		written += ndesc;
		if (unlikely(written > 32)) {
			ring_tx_db(q->adap, &q->q, written);
			written = 0;
		}

		spin_lock(&q->sendq.lock);
		__skb_unlink(skb, &q->sendq);
		if (is_ofld_imm(skb))
			kfree_skb(skb);
	}
	if (likely(written))
		ring_tx_db(q->adap, &q->q, written);
}

/**
 *	ofld_xmit - send a packet through an offload queue
 *	@q: the Tx offload queue
 *	@skb: the packet
 *
 *	Send an offload packet through an SGE offload queue.
 */
static int ofld_xmit(struct sge_ofld_txq *q, struct sk_buff *skb)
{
	skb->priority = calc_tx_flits_ofld(skb);       /* save for restart */
	spin_lock(&q->sendq.lock);
	__skb_queue_tail(&q->sendq, skb);
	if (q->sendq.qlen == 1)
		service_ofldq(q);
	spin_unlock(&q->sendq.lock);
	return NET_XMIT_SUCCESS;
}

/**
 *	restart_ofldq - restart a suspended offload queue
 *	@data: the offload queue to restart
 *
 *	Resumes transmission on a suspended Tx offload queue.
 */
static void restart_ofldq(unsigned long data)
{
	struct sge_ofld_txq *q = (struct sge_ofld_txq *)data;

	spin_lock(&q->sendq.lock);
	q->full = 0;            /* the queue actually is completely empty now */
	service_ofldq(q);
	spin_unlock(&q->sendq.lock);
}

/**
 *	skb_txq - return the Tx queue an offload packet should use
 *	@skb: the packet
 *
 *	Returns the Tx queue an offload packet should use as indicated by bits
 *	1-15 in the packet's queue_mapping.
 */
static inline unsigned int skb_txq(const struct sk_buff *skb)
{
	return skb->queue_mapping >> 1;
}

/**
 *	is_ctrl_pkt - return whether an offload packet is a control packet
 *	@skb: the packet
 *
 *	Returns whether an offload packet should use an OFLD or a CTRL
 *	Tx queue as indicated by bit 0 in the packet's queue_mapping.
 */
static inline unsigned int is_ctrl_pkt(const struct sk_buff *skb)
{
	return skb->queue_mapping & 1;
}

static inline int ofld_send(struct adapter *adap, struct sk_buff *skb)
{
	unsigned int idx = skb_txq(skb);

	if (unlikely(is_ctrl_pkt(skb))) {
		/* Single ctrl queue is a requirement for LE workaround path */
		if (adap->tids.nsftids)
			idx = 0;
		return ctrl_xmit(&adap->sge.ctrlq[idx], skb);
	}
	return ofld_xmit(&adap->sge.ofldtxq[idx], skb);
}

/**
 *	t4_ofld_send - send an offload packet
 *	@adap: the adapter
 *	@skb: the packet
 *
 *	Sends an offload packet.  We use the packet queue_mapping to select the
 *	appropriate Tx queue as follows: bit 0 indicates whether the packet
 *	should be sent as regular or control, bits 1-15 select the queue.
 */
int t4_ofld_send(struct adapter *adap, struct sk_buff *skb)
{
	int ret;

	local_bh_disable();
	ret = ofld_send(adap, skb);
	local_bh_enable();
	return ret;
}

/**
 *	cxgb4_ofld_send - send an offload packet
 *	@dev: the net device
 *	@skb: the packet
 *
 *	Sends an offload packet.  This is an exported version of @t4_ofld_send,
 *	intended for ULDs.
 */
int cxgb4_ofld_send(struct net_device *dev, struct sk_buff *skb)
{
	return t4_ofld_send(netdev2adap(dev), skb);
}
EXPORT_SYMBOL(cxgb4_ofld_send);

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);
}

/**
 *	cxgb4_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.
 */
struct sk_buff *cxgb4_pktgl_to_skb(const struct pkt_gl *gl,
				   unsigned int skb_len, unsigned int pull_len)
{
	struct sk_buff *skb;

	/*
	 * Below we rely on RX_COPY_THRES being less than the smallest Rx buffer
	 * 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) {
		skb = dev_alloc_skb(gl->tot_len);
		if (unlikely(!skb))
			goto out;
		__skb_put(skb, gl->tot_len);
		skb_copy_to_linear_data(skb, gl->va, gl->tot_len);
	} else {
		skb = dev_alloc_skb(skb_len);
		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;
}
EXPORT_SYMBOL(cxgb4_pktgl_to_skb);

/**
 *	t4_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 t4_pktgl_free(const struct pkt_gl *gl)
{
	int n;
	const struct page_frag *p;

	for (p = gl->frags, n = gl->nfrags - 1; n--; p++)
		put_page(p->page);
}

/*
 * Process an MPS trace packet.  Give it an unused protocol number so it won't
 * be delivered to anyone and send it to the stack for capture.
 */
static noinline int handle_trace_pkt(struct adapter *adap,
				     const struct pkt_gl *gl)
{
	struct sk_buff *skb;

	skb = cxgb4_pktgl_to_skb(gl, RX_PULL_LEN, RX_PULL_LEN);
	if (unlikely(!skb)) {
		t4_pktgl_free(gl);
		return 0;
	}

	if (is_t4(adap->params.chip))
		__skb_pull(skb, sizeof(struct cpl_trace_pkt));
	else
		__skb_pull(skb, sizeof(struct cpl_t5_trace_pkt));

	skb_reset_mac_header(skb);
	skb->protocol = htons(0xffff);
	skb->dev = adap->port[0];
	netif_receive_skb(skb);
	return 0;
}

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.adap;
	struct sge *s = &adapter->sge;
	int ret;
	struct sk_buff *skb;

	skb = napi_get_frags(&rxq->rspq.napi);
	if (unlikely(!skb)) {
		t4_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 (rxq->rspq.netdev->features & NETIF_F_RXHASH)
		skb_set_hash(skb, (__force u32)pkt->rsshdr.hash_val,
			     PKT_HASH_TYPE_L3);

	if (unlikely(pkt->vlan_ex)) {
		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(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++;
}

/**
 *	t4_ethrx_handler - process an ingress ethernet packet
 *	@q: the response queue that received the packet
 *	@rsp: the response queue descriptor holding the RX_PKT message
 *	@si: the gather list of packet fragments
 *
 *	Process an ingress ethernet packet and deliver it to the stack.
 */
int t4_ethrx_handler(struct sge_rspq *q, const __be64 *rsp,
		     const struct pkt_gl *si)
{
	bool csum_ok;
	struct sk_buff *skb;
	const struct cpl_rx_pkt *pkt;
	struct sge_eth_rxq *rxq = container_of(q, struct sge_eth_rxq, rspq);
	struct sge *s = &q->adap->sge;
	int cpl_trace_pkt = is_t4(q->adap->params.chip) ?
			    CPL_TRACE_PKT : CPL_TRACE_PKT_T5;

	if (unlikely(*(u8 *)rsp == cpl_trace_pkt))
		return handle_trace_pkt(q->adap, si);

	pkt = (const struct cpl_rx_pkt *)rsp;
	csum_ok = pkt->csum_calc && !pkt->err_vec &&
		  (q->netdev->features & NETIF_F_RXCSUM);
	if ((pkt->l2info & htonl(RXF_TCP)) &&
	    (q->netdev->features & NETIF_F_GRO) && csum_ok && !pkt->ip_frag) {
		do_gro(rxq, si, pkt);
		return 0;
	}

	skb = cxgb4_pktgl_to_skb(si, RX_PKT_SKB_LEN, RX_PULL_LEN);
	if (unlikely(!skb)) {
		t4_pktgl_free(si);
		rxq->stats.rx_drops++;
		return 0;
	}

	__skb_pull(skb, s->pktshift);      /* remove ethernet header padding */
	skb->protocol = eth_type_trans(skb, q->netdev);
	skb_record_rx_queue(skb, q->idx);
	if (skb->dev->features & NETIF_F_RXHASH)
		skb_set_hash(skb, (__force u32)pkt->rsshdr.hash_val,
			     PKT_HASH_TYPE_L3);

	rxq->stats.pkts++;

	if (csum_ok && (pkt->l2info & htonl(RXF_UDP | RXF_TCP))) {
		if (!pkt->ip_frag) {
			skb->ip_summed = CHECKSUM_UNNECESSARY;
			rxq->stats.rx_cso++;
		} else if (pkt->l2info & htonl(RXF_IP)) {
			__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 (unlikely(pkt->vlan_ex)) {
		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(pkt->vlan));
		rxq->stats.vlan_ex++;
	}
	netif_receive_skb(skb);
	return 0;
}

/**
 *	restore_rx_bufs - put back a packet's Rx buffers
 *	@si: the packet gather list
 *	@q: the SGE free list
 *	@frags: number of FL buffers to restore
 *
 *	Puts back on an FL the Rx buffers associated with @si.  The buffers
 *	have already been unmapped and are left unmapped, we mark them so to
 *	prevent further unmapping attempts.
 *
 *	This function undoes a series of @unmap_rx_buf calls when we find out
 *	that the current packet can't be processed right away afterall and we
 *	need to come back to it later.  This is a very rare event and there's
 *	no effort to make this particularly efficient.
 */
static void restore_rx_bufs(const struct pkt_gl *si, struct sge_fl *q,
			    int frags)
{
	struct rx_sw_desc *d;

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

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

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

/**
 *	process_responses - process responses from an SGE response queue
 *	@q: the ingress queue to process
 *	@budget: how many responses can be processed in this round
 *
 *	Process responses from an SGE response queue up to the supplied budget.
 *	Responses include received packets as well as control messages from FW
 *	or HW.
 *
 *	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 *q, int budget)
{
	int ret, rsp_type;
	int budget_left = budget;
	const struct rsp_ctrl *rc;
	struct sge_eth_rxq *rxq = container_of(q, struct sge_eth_rxq, rspq);
	struct adapter *adapter = q->adap;
	struct sge *s = &adapter->sge;

	while (likely(budget_left)) {
		rc = (void *)q->cur_desc + (q->iqe_len - sizeof(*rc));
		if (!is_new_response(rc, q))
			break;

		rmb();
		rsp_type = RSPD_TYPE(rc->type_gen);
		if (likely(rsp_type == RSP_TYPE_FLBUF)) {
			struct page_frag *fp;
			struct pkt_gl si;
			const struct rx_sw_desc *rsd;
			u32 len = ntohl(rc->pldbuflen_qid), bufsz, frags;

			if (len & RSPD_NEWBUF) {
				if (likely(q->offset > 0)) {
					free_rx_bufs(q->adap, &rxq->fl, 1);
					q->offset = 0;
				}
				len = RSPD_LEN(len);
			}
			si.tot_len = len;

			/* gather packet fragments */
			for (frags = 0, fp = si.frags; ; frags++, fp++) {
				rsd = &rxq->fl.sdesc[rxq->fl.cidx];
				bufsz = get_buf_size(adapter, rsd);
				fp->page = rsd->page;
				fp->offset = q->offset;
				fp->size = min(bufsz, len);
				len -= fp->size;
				if (!len)
					break;
				unmap_rx_buf(q->adap, &rxq->fl);
			}

			/*
			 * Last buffer remains mapped so explicitly make it
			 * coherent for CPU access.
			 */
			dma_sync_single_for_cpu(q->adap->pdev_dev,
						get_buf_addr(rsd),
						fp->size, DMA_FROM_DEVICE);

			si.va = page_address(si.frags[0].page) +
				si.frags[0].offset;
			prefetch(si.va);

			si.nfrags = frags + 1;
			ret = q->handler(q, q->cur_desc, &si);
			if (likely(ret == 0))
				q->offset += ALIGN(fp->size, s->fl_align);
			else
				restore_rx_bufs(&si, &rxq->fl, frags);
		} else if (likely(rsp_type == RSP_TYPE_CPL)) {
			ret = q->handler(q, q->cur_desc, NULL);
		} else {
			ret = q->handler(q, (const __be64 *)rc, CXGB4_MSG_AN);
		}

		if (unlikely(ret)) {
			/* couldn't process descriptor, back off for recovery */
			q->next_intr_params = QINTR_TIMER_IDX(NOMEM_TMR_IDX);
			break;
		}

		rspq_next(q);
		budget_left--;
	}

	if (q->offset >= 0 && rxq->fl.size - rxq->fl.avail >= 16)
		__refill_fl(q->adap, &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 params;
	struct sge_rspq *q = container_of(napi, struct sge_rspq, napi);
	int work_done = process_responses(q, budget);
	u32 val;

	if (likely(work_done < budget)) {
		int timer_index;

		napi_complete(napi);
		timer_index = QINTR_TIMER_IDX_GET(q->next_intr_params);

		if (q->adaptive_rx) {
			if (work_done > max(timer_pkt_quota[timer_index],
					    MIN_NAPI_WORK))
				timer_index = (timer_index + 1);
			else
				timer_index = timer_index - 1;

			timer_index = clamp(timer_index, 0, SGE_TIMERREGS - 1);
			q->next_intr_params = QINTR_TIMER_IDX(timer_index) |
							      V_QINTR_CNT_EN;
			params = q->next_intr_params;
		} else {
			params = q->next_intr_params;
			q->next_intr_params = q->intr_params;
		}
	} else
		params = QINTR_TIMER_IDX(7);

	val = CIDXINC(work_done) | SEINTARM(params);
	if (is_t4(q->adap->params.chip)) {
		t4_write_reg(q->adap, MYPF_REG(SGE_PF_GTS),
			     val | INGRESSQID((u32)q->cntxt_id));
	} else {
		writel(val, q->adap->bar2 + q->udb + SGE_UDB_GTS);
		wmb();
	}
	return work_done;
}

/*
 * The MSI-X interrupt handler for an SGE response queue.
 */
irqreturn_t t4_sge_intr_msix(int irq, void *cookie)
{
	struct sge_rspq *q = cookie;

	napi_schedule(&q->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 *adap)
{
	unsigned int credits;
	const struct rsp_ctrl *rc;
	struct sge_rspq *q = &adap->sge.intrq;
	u32 val;

	spin_lock(&adap->sge.intrq_lock);
	for (credits = 0; ; credits++) {
		rc = (void *)q->cur_desc + (q->iqe_len - sizeof(*rc));
		if (!is_new_response(rc, q))
			break;

		rmb();
		if (RSPD_TYPE(rc->type_gen) == RSP_TYPE_INTR) {
			unsigned int qid = ntohl(rc->pldbuflen_qid);

			qid -= adap->sge.ingr_start;
			napi_schedule(&adap->sge.ingr_map[qid]->napi);
		}

		rspq_next(q);
	}

	val =  CIDXINC(credits) | SEINTARM(q->intr_params);
	if (is_t4(adap->params.chip)) {
		t4_write_reg(adap, MYPF_REG(SGE_PF_GTS),
			     val | INGRESSQID(q->cntxt_id));
	} else {
		writel(val, adap->bar2 + q->udb + SGE_UDB_GTS);
		wmb();
	}
	spin_unlock(&adap->sge.intrq_lock);
	return credits;
}

/*
 * The MSI interrupt handler, which 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 t4_intr_msi(int irq, void *cookie)
{
	struct adapter *adap = cookie;

	t4_slow_intr_handler(adap);
	process_intrq(adap);
	return IRQ_HANDLED;
}

/*
 * Interrupt handler for legacy INTx interrupts.
 * Handles data events from SGE response queues as well as error and other
 * async events as they all use the same interrupt line.
 */
static irqreturn_t t4_intr_intx(int irq, void *cookie)
{
	struct adapter *adap = cookie;

	t4_write_reg(adap, MYPF_REG(PCIE_PF_CLI), 0);
	if (t4_slow_intr_handler(adap) | process_intrq(adap))
		return IRQ_HANDLED;
	return IRQ_NONE;             /* probably shared interrupt */
}

/**
 *	t4_intr_handler - select the top-level interrupt handler
 *	@adap: the adapter
 *
 *	Selects the top-level interrupt handler based on the type of interrupts
 *	(MSI-X, MSI, or INTx).
 */
irq_handler_t t4_intr_handler(struct adapter *adap)
{
	if (adap->flags & USING_MSIX)
		return t4_sge_intr_msix;
	if (adap->flags & USING_MSI)
		return t4_intr_msi;
	return t4_intr_intx;
}

static void sge_rx_timer_cb(unsigned long data)
{
	unsigned long m;
	unsigned int i, idma_same_state_cnt[2];
	struct adapter *adap = (struct adapter *)data;
	struct sge *s = &adap->sge;

	for (i = 0; i < ARRAY_SIZE(s->starving_fl); i++)
		for (m = s->starving_fl[i]; m; m &= m - 1) {
			struct sge_eth_rxq *rxq;
			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();

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

	t4_write_reg(adap, SGE_DEBUG_INDEX, 13);
	idma_same_state_cnt[0] = t4_read_reg(adap, SGE_DEBUG_DATA_HIGH);
	idma_same_state_cnt[1] = t4_read_reg(adap, SGE_DEBUG_DATA_LOW);

	for (i = 0; i < 2; i++) {
		u32 debug0, debug11;

		/* If the Ingress DMA Same State Counter ("timer") is less
		 * than 1s, then we can reset our synthesized Stall Timer and
		 * continue.  If we have previously emitted warnings about a
		 * potential stalled Ingress Queue, issue a note indicating
		 * that the Ingress Queue has resumed forward progress.
		 */
		if (idma_same_state_cnt[i] < s->idma_1s_thresh) {
			if (s->idma_stalled[i] >= SGE_IDMA_WARN_THRESH)
				CH_WARN(adap, "SGE idma%d, queue%u,resumed after %d sec\n",
					i, s->idma_qid[i],
					s->idma_stalled[i]/HZ);
			s->idma_stalled[i] = 0;
			continue;
		}

		/* Synthesize an SGE Ingress DMA Same State Timer in the Hz
		 * domain.  The first time we get here it'll be because we
		 * passed the 1s Threshold; each additional time it'll be
		 * because the RX Timer Callback is being fired on its regular
		 * schedule.
		 *
		 * If the stall is below our Potential Hung Ingress Queue
		 * Warning Threshold, continue.
		 */
		if (s->idma_stalled[i] == 0)
			s->idma_stalled[i] = HZ;
		else
			s->idma_stalled[i] += RX_QCHECK_PERIOD;

		if (s->idma_stalled[i] < SGE_IDMA_WARN_THRESH)
			continue;

		/* We'll issue a warning every SGE_IDMA_WARN_REPEAT Hz */
		if (((s->idma_stalled[i] - HZ) % SGE_IDMA_WARN_REPEAT) != 0)
			continue;

		/* Read and save the SGE IDMA State and Queue ID information.
		 * We do this every time in case it changes across time ...
		 */
		t4_write_reg(adap, SGE_DEBUG_INDEX, 0);
		debug0 = t4_read_reg(adap, SGE_DEBUG_DATA_LOW);
		s->idma_state[i] = (debug0 >> (i * 9)) & 0x3f;

		t4_write_reg(adap, SGE_DEBUG_INDEX, 11);
		debug11 = t4_read_reg(adap, SGE_DEBUG_DATA_LOW);
		s->idma_qid[i] = (debug11 >> (i * 16)) & 0xffff;

		CH_WARN(adap, "SGE idma%u, queue%u, maybe stuck state%u %dsecs (debug0=%#x, debug11=%#x)\n",
			i, s->idma_qid[i], s->idma_state[i],
			s->idma_stalled[i]/HZ, debug0, debug11);
		t4_sge_decode_idma_state(adap, s->idma_state[i]);
	}

	mod_timer(&s->rx_timer, jiffies + RX_QCHECK_PERIOD);
}

static void sge_tx_timer_cb(unsigned long data)
{
	unsigned long m;
	unsigned int i, budget;
	struct adapter *adap = (struct adapter *)data;
	struct sge *s = &adap->sge;

	for (i = 0; i < ARRAY_SIZE(s->txq_maperr); i++)
		for (m = s->txq_maperr[i]; m; m &= m - 1) {
			unsigned long id = __ffs(m) + i * BITS_PER_LONG;
			struct sge_ofld_txq *txq = s->egr_map[id];

			clear_bit(id, s->txq_maperr);
			tasklet_schedule(&txq->qresume_tsk);
		}

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

		if (q->q.in_use &&
		    time_after_eq(jiffies, q->txq->trans_start + HZ / 100) &&
		    __netif_tx_trylock(q->txq)) {
			int avail = reclaimable(&q->q);

			if (avail) {
				if (avail > budget)
					avail = budget;

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

		if (++i >= s->ethqsets)
			i = 0;
	} while (budget && i != s->ethtxq_rover);
	s->ethtxq_rover = i;
	mod_timer(&s->tx_timer, jiffies + (budget ? TX_QCHECK_PERIOD : 2));
}

/**
 *      udb_address - return the BAR2 User Doorbell address for a Queue
 *      @adap: the adapter
 *      @cntxt_id: the Queue Context ID
 *      @qpp: Queues Per Page (for all PFs)
 *
 *      Returns the BAR2 address of the user Doorbell associated with the
 *      indicated Queue Context ID.  Note that this is only applicable
 *      for T5 and later.
 */
static u64 udb_address(struct adapter *adap, unsigned int cntxt_id,
		       unsigned int qpp)
{
	u64 udb;
	unsigned int s_qpp;
	unsigned short udb_density;
	unsigned long qpshift;
	int page;

	BUG_ON(is_t4(adap->params.chip));

	s_qpp = (QUEUESPERPAGEPF0 +
		(QUEUESPERPAGEPF1 - QUEUESPERPAGEPF0) * adap->fn);
	udb_density = 1 << ((qpp >> s_qpp) & QUEUESPERPAGEPF0_MASK);
	qpshift = PAGE_SHIFT - ilog2(udb_density);
	udb = (u64)cntxt_id << qpshift;
	udb &= PAGE_MASK;
	page = udb / PAGE_SIZE;
	udb += (cntxt_id - (page * udb_density)) * SGE_UDB_SIZE;

	return udb;
}

static u64 udb_address_eq(struct adapter *adap, unsigned int cntxt_id)
{
	return udb_address(adap, cntxt_id,
			   t4_read_reg(adap, SGE_EGRESS_QUEUES_PER_PAGE_PF));
}

static u64 udb_address_iq(struct adapter *adap, unsigned int cntxt_id)
{
	return udb_address(adap, cntxt_id,
			   t4_read_reg(adap, SGE_INGRESS_QUEUES_PER_PAGE_PF));
}

int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq,
		     struct net_device *dev, int intr_idx,
		     struct sge_fl *fl, rspq_handler_t hnd)
{
	int ret, flsz = 0;
	struct fw_iq_cmd c;
	struct sge *s = &adap->sge;
	struct port_info *pi = netdev_priv(dev);

	/* Size needs to be multiple of 16, including status entry. */
	iq->size = roundup(iq->size, 16);

	iq->desc = alloc_ring(adap->pdev_dev, iq->size, iq->iqe_len, 0,
			      &iq->phys_addr, NULL, 0, NUMA_NO_NODE);
	if (!iq->desc)
		return -ENOMEM;

	memset(&c, 0, sizeof(c));
	c.op_to_vfn = htonl(FW_CMD_OP(FW_IQ_CMD) | FW_CMD_REQUEST |
			    FW_CMD_WRITE | FW_CMD_EXEC |
			    FW_IQ_CMD_PFN(adap->fn) | FW_IQ_CMD_VFN(0));
	c.alloc_to_len16 = htonl(FW_IQ_CMD_ALLOC | FW_IQ_CMD_IQSTART(1) |
				 FW_LEN16(c));
	c.type_to_iqandstindex = htonl(FW_IQ_CMD_TYPE(FW_IQ_TYPE_FL_INT_CAP) |
		FW_IQ_CMD_IQASYNCH(fwevtq) | FW_IQ_CMD_VIID(pi->viid) |
		FW_IQ_CMD_IQANDST(intr_idx < 0) | FW_IQ_CMD_IQANUD(1) |
		FW_IQ_CMD_IQANDSTINDEX(intr_idx >= 0 ? intr_idx :
							-intr_idx - 1));
	c.iqdroprss_to_iqesize = htons(FW_IQ_CMD_IQPCIECH(pi->tx_chan) |
		FW_IQ_CMD_IQGTSMODE |
		FW_IQ_CMD_IQINTCNTTHRESH(iq->pktcnt_idx) |
		FW_IQ_CMD_IQESIZE(ilog2(iq->iqe_len) - 4));
	c.iqsize = htons(iq->size);
	c.iqaddr = cpu_to_be64(iq->phys_addr);

	if (fl) {
		fl->size = roundup(fl->size, 8);
		fl->desc = alloc_ring(adap->pdev_dev, fl->size, sizeof(__be64),
				      sizeof(struct rx_sw_desc), &fl->addr,
				      &fl->sdesc, s->stat_len, NUMA_NO_NODE);
		if (!fl->desc)
			goto fl_nomem;

		flsz = fl->size / 8 + s->stat_len / sizeof(struct tx_desc);
		c.iqns_to_fl0congen = htonl(FW_IQ_CMD_FL0PACKEN(1) |
					    FW_IQ_CMD_FL0FETCHRO(1) |
					    FW_IQ_CMD_FL0DATARO(1) |
					    FW_IQ_CMD_FL0PADEN(1));
		c.fl0dcaen_to_fl0cidxfthresh = htons(FW_IQ_CMD_FL0FBMIN(2) |
				FW_IQ_CMD_FL0FBMAX(3));
		c.fl0size = htons(flsz);
		c.fl0addr = cpu_to_be64(fl->addr);
	}

	ret = t4_wr_mbox(adap, adap->fn, &c, sizeof(c), &c);
	if (ret)
		goto err;

	netif_napi_add(dev, &iq->napi, napi_rx_handler, 64);
	iq->cur_desc = iq->desc;
	iq->cidx = 0;
	iq->gen = 1;
	iq->next_intr_params = iq->intr_params;
	iq->cntxt_id = ntohs(c.iqid);
	iq->abs_id = ntohs(c.physiqid);
	if (!is_t4(adap->params.chip))
		iq->udb = udb_address_iq(adap, iq->cntxt_id);
	iq->size--;                           /* subtract status entry */
	iq->netdev = dev;
	iq->handler = hnd;

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

	adap->sge.ingr_map[iq->cntxt_id - adap->sge.ingr_start] = iq;

	if (fl) {
		fl->cntxt_id = ntohs(c.fl0id);
		fl->avail = fl->pend_cred = 0;
		fl->pidx = fl->cidx = 0;
		fl->alloc_failed = fl->large_alloc_failed = fl->starving = 0;
		adap->sge.egr_map[fl->cntxt_id - adap->sge.egr_start] = fl;

		/* Note, we must initialize the Free List User Doorbell
		 * address before refilling the Free List!
		 */
		if (!is_t4(adap->params.chip))
			fl->udb = udb_address_eq(adap, fl->cntxt_id);
		refill_fl(adap, fl, fl_cap(fl), GFP_KERNEL);
	}
	return 0;

fl_nomem:
	ret = -ENOMEM;
err:
	if (iq->desc) {
		dma_free_coherent(adap->pdev_dev, iq->size * iq->iqe_len,
				  iq->desc, iq->phys_addr);
		iq->desc = NULL;
	}
	if (fl && fl->desc) {
		kfree(fl->sdesc);
		fl->sdesc = NULL;
		dma_free_coherent(adap->pdev_dev, flsz * sizeof(struct tx_desc),
				  fl->desc, fl->addr);
		fl->desc = NULL;
	}
	return ret;
}

static void init_txq(struct adapter *adap, struct sge_txq *q, unsigned int id)
{
	q->cntxt_id = id;
	if (!is_t4(adap->params.chip))
		q->udb = udb_address_eq(adap, q->cntxt_id);

	q->in_use = 0;
	q->cidx = q->pidx = 0;
	q->stops = q->restarts = 0;
	q->stat = (void *)&q->desc[q->size];
	spin_lock_init(&q->db_lock);
	adap->sge.egr_map[id - adap->sge.egr_start] = q;
}

int t4_sge_alloc_eth_txq(struct adapter *adap, struct sge_eth_txq *txq,
			 struct net_device *dev, struct netdev_queue *netdevq,
			 unsigned int iqid)
{
	int ret, nentries;
	struct fw_eq_eth_cmd c;
	struct sge *s = &adap->sge;
	struct port_info *pi = netdev_priv(dev);

	/* Add status entries */
	nentries = txq->q.size + s->stat_len / sizeof(struct tx_desc);

	txq->q.desc = alloc_ring(adap->pdev_dev, txq->q.size,
			sizeof(struct tx_desc), sizeof(struct tx_sw_desc),
			&txq->q.phys_addr, &txq->q.sdesc, s->stat_len,
			netdev_queue_numa_node_read(netdevq));
	if (!txq->q.desc)
		return -ENOMEM;

	memset(&c, 0, sizeof(c));
	c.op_to_vfn = htonl(FW_CMD_OP(FW_EQ_ETH_CMD) | FW_CMD_REQUEST |
			    FW_CMD_WRITE | FW_CMD_EXEC |
			    FW_EQ_ETH_CMD_PFN(adap->fn) | FW_EQ_ETH_CMD_VFN(0));
	c.alloc_to_len16 = htonl(FW_EQ_ETH_CMD_ALLOC |
				 FW_EQ_ETH_CMD_EQSTART | FW_LEN16(c));
	c.viid_pkd = htonl(FW_EQ_ETH_CMD_AUTOEQUEQE |
			   FW_EQ_ETH_CMD_VIID(pi->viid));
	c.fetchszm_to_iqid = htonl(FW_EQ_ETH_CMD_HOSTFCMODE(2) |
				   FW_EQ_ETH_CMD_PCIECHN(pi->tx_chan) |
				   FW_EQ_ETH_CMD_FETCHRO(1) |
				   FW_EQ_ETH_CMD_IQID(iqid));
	c.dcaen_to_eqsize = htonl(FW_EQ_ETH_CMD_FBMIN(2) |
				  FW_EQ_ETH_CMD_FBMAX(3) |
				  FW_EQ_ETH_CMD_CIDXFTHRESH(5) |
				  FW_EQ_ETH_CMD_EQSIZE(nentries));
	c.eqaddr = cpu_to_be64(txq->q.phys_addr);

	ret = t4_wr_mbox(adap, adap->fn, &c, sizeof(c), &c);
	if (ret) {
		kfree(txq->q.sdesc);
		txq->q.sdesc = NULL;
		dma_free_coherent(adap->pdev_dev,
				  nentries * sizeof(struct tx_desc),
				  txq->q.desc, txq->q.phys_addr);
		txq->q.desc = NULL;
		return ret;
	}

	init_txq(adap, &txq->q, FW_EQ_ETH_CMD_EQID_GET(ntohl(c.eqid_pkd)));
	txq->txq = netdevq;
	txq->tso = txq->tx_cso = txq->vlan_ins = 0;
	txq->mapping_err = 0;
	return 0;
}

int t4_sge_alloc_ctrl_txq(struct adapter *adap, struct sge_ctrl_txq *txq,
			  struct net_device *dev, unsigned int iqid,
			  unsigned int cmplqid)
{
	int ret, nentries;
	struct fw_eq_ctrl_cmd c;
	struct sge *s = &adap->sge;
	struct port_info *pi = netdev_priv(dev);

	/* Add status entries */
	nentries = txq->q.size + s->stat_len / sizeof(struct tx_desc);

	txq->q.desc = alloc_ring(adap->pdev_dev, nentries,
				 sizeof(struct tx_desc), 0, &txq->q.phys_addr,
				 NULL, 0, NUMA_NO_NODE);
	if (!txq->q.desc)
		return -ENOMEM;

	c.op_to_vfn = htonl(FW_CMD_OP(FW_EQ_CTRL_CMD) | FW_CMD_REQUEST |
			    FW_CMD_WRITE | FW_CMD_EXEC |
			    FW_EQ_CTRL_CMD_PFN(adap->fn) |
			    FW_EQ_CTRL_CMD_VFN(0));
	c.alloc_to_len16 = htonl(FW_EQ_CTRL_CMD_ALLOC |
				 FW_EQ_CTRL_CMD_EQSTART | FW_LEN16(c));
	c.cmpliqid_eqid = htonl(FW_EQ_CTRL_CMD_CMPLIQID(cmplqid));
	c.physeqid_pkd = htonl(0);
	c.fetchszm_to_iqid = htonl(FW_EQ_CTRL_CMD_HOSTFCMODE(2) |
				   FW_EQ_CTRL_CMD_PCIECHN(pi->tx_chan) |
				   FW_EQ_CTRL_CMD_FETCHRO |
				   FW_EQ_CTRL_CMD_IQID(iqid));
	c.dcaen_to_eqsize = htonl(FW_EQ_CTRL_CMD_FBMIN(2) |
				  FW_EQ_CTRL_CMD_FBMAX(3) |
				  FW_EQ_CTRL_CMD_CIDXFTHRESH(5) |
				  FW_EQ_CTRL_CMD_EQSIZE(nentries));
	c.eqaddr = cpu_to_be64(txq->q.phys_addr);

	ret = t4_wr_mbox(adap, adap->fn, &c, sizeof(c), &c);
	if (ret) {
		dma_free_coherent(adap->pdev_dev,
				  nentries * sizeof(struct tx_desc),
				  txq->q.desc, txq->q.phys_addr);
		txq->q.desc = NULL;
		return ret;
	}

	init_txq(adap, &txq->q, FW_EQ_CTRL_CMD_EQID_GET(ntohl(c.cmpliqid_eqid)));
	txq->adap = adap;
	skb_queue_head_init(&txq->sendq);
	tasklet_init(&txq->qresume_tsk, restart_ctrlq, (unsigned long)txq);
	txq->full = 0;
	return 0;
}

int t4_sge_alloc_ofld_txq(struct adapter *adap, struct sge_ofld_txq *txq,
			  struct net_device *dev, unsigned int iqid)
{
	int ret, nentries;
	struct fw_eq_ofld_cmd c;
	struct sge *s = &adap->sge;
	struct port_info *pi = netdev_priv(dev);

	/* Add status entries */
	nentries = txq->q.size + s->stat_len / sizeof(struct tx_desc);

	txq->q.desc = alloc_ring(adap->pdev_dev, txq->q.size,
			sizeof(struct tx_desc), sizeof(struct tx_sw_desc),
			&txq->q.phys_addr, &txq->q.sdesc, s->stat_len,
			NUMA_NO_NODE);
	if (!txq->q.desc)
		return -ENOMEM;

	memset(&c, 0, sizeof(c));
	c.op_to_vfn = htonl(FW_CMD_OP(FW_EQ_OFLD_CMD) | FW_CMD_REQUEST |
			    FW_CMD_WRITE | FW_CMD_EXEC |
			    FW_EQ_OFLD_CMD_PFN(adap->fn) |
			    FW_EQ_OFLD_CMD_VFN(0));
	c.alloc_to_len16 = htonl(FW_EQ_OFLD_CMD_ALLOC |
				 FW_EQ_OFLD_CMD_EQSTART | FW_LEN16(c));
	c.fetchszm_to_iqid = htonl(FW_EQ_OFLD_CMD_HOSTFCMODE(2) |
				   FW_EQ_OFLD_CMD_PCIECHN(pi->tx_chan) |
				   FW_EQ_OFLD_CMD_FETCHRO(1) |
				   FW_EQ_OFLD_CMD_IQID(iqid));
	c.dcaen_to_eqsize = htonl(FW_EQ_OFLD_CMD_FBMIN(2) |
				  FW_EQ_OFLD_CMD_FBMAX(3) |
				  FW_EQ_OFLD_CMD_CIDXFTHRESH(5) |
				  FW_EQ_OFLD_CMD_EQSIZE(nentries));
	c.eqaddr = cpu_to_be64(txq->q.phys_addr);

	ret = t4_wr_mbox(adap, adap->fn, &c, sizeof(c), &c);
	if (ret) {
		kfree(txq->q.sdesc);
		txq->q.sdesc = NULL;
		dma_free_coherent(adap->pdev_dev,
				  nentries * sizeof(struct tx_desc),
				  txq->q.desc, txq->q.phys_addr);
		txq->q.desc = NULL;
		return ret;
	}

	init_txq(adap, &txq->q, FW_EQ_OFLD_CMD_EQID_GET(ntohl(c.eqid_pkd)));
	txq->adap = adap;
	skb_queue_head_init(&txq->sendq);
	tasklet_init(&txq->qresume_tsk, restart_ofldq, (unsigned long)txq);
	txq->full = 0;
	txq->mapping_err = 0;
	return 0;
}

static void free_txq(struct adapter *adap, struct sge_txq *q)
{
	struct sge *s = &adap->sge;

	dma_free_coherent(adap->pdev_dev,
			  q->size * sizeof(struct tx_desc) + s->stat_len,
			  q->desc, q->phys_addr);
	q->cntxt_id = 0;
	q->sdesc = NULL;
	q->desc = NULL;
}

static void free_rspq_fl(struct adapter *adap, struct sge_rspq *rq,
			 struct sge_fl *fl)
{
	struct sge *s = &adap->sge;
	unsigned int fl_id = fl ? fl->cntxt_id : 0xffff;

	adap->sge.ingr_map[rq->cntxt_id - adap->sge.ingr_start] = NULL;
	t4_iq_free(adap, adap->fn, adap->fn, 0, FW_IQ_TYPE_FL_INT_CAP,
		   rq->cntxt_id, fl_id, 0xffff);
	dma_free_coherent(adap->pdev_dev, (rq->size + 1) * rq->iqe_len,
			  rq->desc, rq->phys_addr);
	netif_napi_del(&rq->napi);
	rq->netdev = NULL;
	rq->cntxt_id = rq->abs_id = 0;
	rq->desc = NULL;

	if (fl) {
		free_rx_bufs(adap, fl, fl->avail);
		dma_free_coherent(adap->pdev_dev, fl->size * 8 + s->stat_len,
				  fl->desc, fl->addr);
		kfree(fl->sdesc);
		fl->sdesc = NULL;
		fl->cntxt_id = 0;
		fl->desc = NULL;
	}
}

/**
 *      t4_free_ofld_rxqs - free a block of consecutive Rx queues
 *      @adap: the adapter
 *      @n: number of queues
 *      @q: pointer to first queue
 *
 *      Release the resources of a consecutive block of offload Rx queues.
 */
void t4_free_ofld_rxqs(struct adapter *adap, int n, struct sge_ofld_rxq *q)
{
	for ( ; n; n--, q++)
		if (q->rspq.desc)
			free_rspq_fl(adap, &q->rspq,
				     q->fl.size ? &q->fl : NULL);
}

/**
 *	t4_free_sge_resources - free SGE resources
 *	@adap: the adapter
 *
 *	Frees resources used by the SGE queue sets.
 */
void t4_free_sge_resources(struct adapter *adap)
{
	int i;
	struct sge_eth_rxq *eq = adap->sge.ethrxq;
	struct sge_eth_txq *etq = adap->sge.ethtxq;

	/* clean up Ethernet Tx/Rx queues */
	for (i = 0; i < adap->sge.ethqsets; i++, eq++, etq++) {
		if (eq->rspq.desc)
			free_rspq_fl(adap, &eq->rspq,
				     eq->fl.size ? &eq->fl : NULL);
		if (etq->q.desc) {
			t4_eth_eq_free(adap, adap->fn, adap->fn, 0,
				       etq->q.cntxt_id);
			free_tx_desc(adap, &etq->q, etq->q.in_use, true);
			kfree(etq->q.sdesc);
			free_txq(adap, &etq->q);
		}
	}

	/* clean up RDMA and iSCSI Rx queues */
	t4_free_ofld_rxqs(adap, adap->sge.ofldqsets, adap->sge.ofldrxq);
	t4_free_ofld_rxqs(adap, adap->sge.rdmaqs, adap->sge.rdmarxq);
	t4_free_ofld_rxqs(adap, adap->sge.rdmaciqs, adap->sge.rdmaciq);

	/* clean up offload Tx queues */
	for (i = 0; i < ARRAY_SIZE(adap->sge.ofldtxq); i++) {
		struct sge_ofld_txq *q = &adap->sge.ofldtxq[i];

		if (q->q.desc) {
			tasklet_kill(&q->qresume_tsk);
			t4_ofld_eq_free(adap, adap->fn, adap->fn, 0,
					q->q.cntxt_id);
			free_tx_desc(adap, &q->q, q->q.in_use, false);
			kfree(q->q.sdesc);
			__skb_queue_purge(&q->sendq);
			free_txq(adap, &q->q);
		}
	}

	/* clean up control Tx queues */
	for (i = 0; i < ARRAY_SIZE(adap->sge.ctrlq); i++) {
		struct sge_ctrl_txq *cq = &adap->sge.ctrlq[i];

		if (cq->q.desc) {
			tasklet_kill(&cq->qresume_tsk);
			t4_ctrl_eq_free(adap, adap->fn, adap->fn, 0,
					cq->q.cntxt_id);
			__skb_queue_purge(&cq->sendq);
			free_txq(adap, &cq->q);
		}
	}

	if (adap->sge.fw_evtq.desc)
		free_rspq_fl(adap, &adap->sge.fw_evtq, NULL);

	if (adap->sge.intrq.desc)
		free_rspq_fl(adap, &adap->sge.intrq, NULL);

	/* clear the reverse egress queue map */
	memset(adap->sge.egr_map, 0, sizeof(adap->sge.egr_map));
}

void t4_sge_start(struct adapter *adap)
{
	adap->sge.ethtxq_rover = 0;
	mod_timer(&adap->sge.rx_timer, jiffies + RX_QCHECK_PERIOD);
	mod_timer(&adap->sge.tx_timer, jiffies + TX_QCHECK_PERIOD);
}

/**
 *	t4_sge_stop - disable SGE operation
 *	@adap: 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 t4_sge_stop(struct adapter *adap)
{
	int i;
	struct sge *s = &adap->sge;

	if (in_interrupt())  /* actions below require waiting */
		return;

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

	for (i = 0; i < ARRAY_SIZE(s->ofldtxq); i++) {
		struct sge_ofld_txq *q = &s->ofldtxq[i];

		if (q->q.desc)
			tasklet_kill(&q->qresume_tsk);
	}
	for (i = 0; i < ARRAY_SIZE(s->ctrlq); i++) {
		struct sge_ctrl_txq *cq = &s->ctrlq[i];

		if (cq->q.desc)
			tasklet_kill(&cq->qresume_tsk);
	}
}

/**
 *	t4_sge_init - initialize SGE
 *	@adap: the adapter
 *
 *	Performs SGE initialization needed every time after a chip reset.
 *	We do not initialize any of the queues here, instead the driver
 *	top-level must request them individually.
 *
 *	Called in two different modes:
 *
 *	 1. Perform actual hardware initialization and record hard-coded
 *	    parameters which were used.  This gets used when we're the
 *	    Master PF and the Firmware Configuration File support didn't
 *	    work for some reason.
 *
 *	 2. We're not the Master PF or initialization was performed with
 *	    a Firmware Configuration File.  In this case we need to grab
 *	    any of the SGE operating parameters that we need to have in
 *	    order to do our job and make sure we can live with them ...
 */

static int t4_sge_init_soft(struct adapter *adap)
{
	struct sge *s = &adap->sge;
	u32 fl_small_pg, fl_large_pg, fl_small_mtu, fl_large_mtu;
	u32 timer_value_0_and_1, timer_value_2_and_3, timer_value_4_and_5;
	u32 ingress_rx_threshold;

	/*
	 * Verify that CPL messages are going to the Ingress Queue for
	 * process_responses() and that only packet data is going to the
	 * Free Lists.
	 */
	if ((t4_read_reg(adap, SGE_CONTROL) & RXPKTCPLMODE_MASK) !=
	    RXPKTCPLMODE(X_RXPKTCPLMODE_SPLIT)) {
		dev_err(adap->pdev_dev, "bad SGE CPL MODE\n");
		return -EINVAL;
	}

	/*
	 * Validate the Host Buffer Register Array indices that we want to
	 * use ...
	 *
	 * XXX Note that we should really read through the Host Buffer Size
	 * XXX register array and find the indices of the Buffer Sizes which
	 * XXX meet our needs!
	 */
	#define READ_FL_BUF(x) \
		t4_read_reg(adap, SGE_FL_BUFFER_SIZE0+(x)*sizeof(u32))

	fl_small_pg = READ_FL_BUF(RX_SMALL_PG_BUF);
	fl_large_pg = READ_FL_BUF(RX_LARGE_PG_BUF);
	fl_small_mtu = READ_FL_BUF(RX_SMALL_MTU_BUF);
	fl_large_mtu = READ_FL_BUF(RX_LARGE_MTU_BUF);

	/* We only bother using the Large Page logic if the Large Page Buffer
	 * is larger than our Page Size Buffer.
	 */
	if (fl_large_pg <= fl_small_pg)
		fl_large_pg = 0;

	#undef READ_FL_BUF

	/* The Page Size Buffer must be exactly equal to our Page Size and the
	 * Large Page Size Buffer should be 0 (per above) or a power of 2.
	 */
	if (fl_small_pg != PAGE_SIZE ||
	    (fl_large_pg & (fl_large_pg-1)) != 0) {
		dev_err(adap->pdev_dev, "bad SGE FL page buffer sizes [%d, %d]\n",
			fl_small_pg, fl_large_pg);
		return -EINVAL;
	}
	if (fl_large_pg)
		s->fl_pg_order = ilog2(fl_large_pg) - PAGE_SHIFT;

	if (fl_small_mtu < FL_MTU_SMALL_BUFSIZE(adap) ||
	    fl_large_mtu < FL_MTU_LARGE_BUFSIZE(adap)) {
		dev_err(adap->pdev_dev, "bad SGE FL MTU sizes [%d, %d]\n",
			fl_small_mtu, fl_large_mtu);
		return -EINVAL;
	}

	/*
	 * Retrieve our RX interrupt holdoff timer values and counter
	 * threshold values from the SGE parameters.
	 */
	timer_value_0_and_1 = t4_read_reg(adap, SGE_TIMER_VALUE_0_AND_1);
	timer_value_2_and_3 = t4_read_reg(adap, SGE_TIMER_VALUE_2_AND_3);
	timer_value_4_and_5 = t4_read_reg(adap, SGE_TIMER_VALUE_4_AND_5);
	s->timer_val[0] = core_ticks_to_us(adap,
		TIMERVALUE0_GET(timer_value_0_and_1));
	s->timer_val[1] = core_ticks_to_us(adap,
		TIMERVALUE1_GET(timer_value_0_and_1));
	s->timer_val[2] = core_ticks_to_us(adap,
		TIMERVALUE2_GET(timer_value_2_and_3));
	s->timer_val[3] = core_ticks_to_us(adap,
		TIMERVALUE3_GET(timer_value_2_and_3));
	s->timer_val[4] = core_ticks_to_us(adap,
		TIMERVALUE4_GET(timer_value_4_and_5));
	s->timer_val[5] = core_ticks_to_us(adap,
		TIMERVALUE5_GET(timer_value_4_and_5));

	ingress_rx_threshold = t4_read_reg(adap, SGE_INGRESS_RX_THRESHOLD);
	s->counter_val[0] = THRESHOLD_0_GET(ingress_rx_threshold);
	s->counter_val[1] = THRESHOLD_1_GET(ingress_rx_threshold);
	s->counter_val[2] = THRESHOLD_2_GET(ingress_rx_threshold);
	s->counter_val[3] = THRESHOLD_3_GET(ingress_rx_threshold);

	return 0;
}

static int t4_sge_init_hard(struct adapter *adap)
{
	struct sge *s = &adap->sge;

	/*
	 * Set up our basic SGE mode to deliver CPL messages to our Ingress
	 * Queue and Packet Date to the Free List.
	 */
	t4_set_reg_field(adap, SGE_CONTROL, RXPKTCPLMODE_MASK,
			 RXPKTCPLMODE_MASK);

	/*
	 * Set up to drop DOORBELL writes when the DOORBELL FIFO overflows
	 * and generate an interrupt when this occurs so we can recover.
	 */
	if (is_t4(adap->params.chip)) {
		t4_set_reg_field(adap, A_SGE_DBFIFO_STATUS,
				 V_HP_INT_THRESH(M_HP_INT_THRESH) |
				 V_LP_INT_THRESH(M_LP_INT_THRESH),
				 V_HP_INT_THRESH(dbfifo_int_thresh) |
				 V_LP_INT_THRESH(dbfifo_int_thresh));
	} else {
		t4_set_reg_field(adap, A_SGE_DBFIFO_STATUS,
				 V_LP_INT_THRESH_T5(M_LP_INT_THRESH_T5),
				 V_LP_INT_THRESH_T5(dbfifo_int_thresh));
		t4_set_reg_field(adap, SGE_DBFIFO_STATUS2,
				 V_HP_INT_THRESH_T5(M_HP_INT_THRESH_T5),
				 V_HP_INT_THRESH_T5(dbfifo_int_thresh));
	}
	t4_set_reg_field(adap, A_SGE_DOORBELL_CONTROL, F_ENABLE_DROP,
			F_ENABLE_DROP);

	/*
	 * SGE_FL_BUFFER_SIZE0 (RX_SMALL_PG_BUF) is set up by
	 * t4_fixup_host_params().
	 */
	s->fl_pg_order = FL_PG_ORDER;
	if (s->fl_pg_order)
		t4_write_reg(adap,
			     SGE_FL_BUFFER_SIZE0+RX_LARGE_PG_BUF*sizeof(u32),
			     PAGE_SIZE << FL_PG_ORDER);
	t4_write_reg(adap, SGE_FL_BUFFER_SIZE0+RX_SMALL_MTU_BUF*sizeof(u32),
		     FL_MTU_SMALL_BUFSIZE(adap));
	t4_write_reg(adap, SGE_FL_BUFFER_SIZE0+RX_LARGE_MTU_BUF*sizeof(u32),
		     FL_MTU_LARGE_BUFSIZE(adap));

	/*
	 * Note that the SGE Ingress Packet Count Interrupt Threshold and
	 * Timer Holdoff values must be supplied by our caller.
	 */
	t4_write_reg(adap, SGE_INGRESS_RX_THRESHOLD,
		     THRESHOLD_0(s->counter_val[0]) |
		     THRESHOLD_1(s->counter_val[1]) |
		     THRESHOLD_2(s->counter_val[2]) |
		     THRESHOLD_3(s->counter_val[3]));
	t4_write_reg(adap, SGE_TIMER_VALUE_0_AND_1,
		     TIMERVALUE0(us_to_core_ticks(adap, s->timer_val[0])) |
		     TIMERVALUE1(us_to_core_ticks(adap, s->timer_val[1])));
	t4_write_reg(adap, SGE_TIMER_VALUE_2_AND_3,
		     TIMERVALUE2(us_to_core_ticks(adap, s->timer_val[2])) |
		     TIMERVALUE3(us_to_core_ticks(adap, s->timer_val[3])));
	t4_write_reg(adap, SGE_TIMER_VALUE_4_AND_5,
		     TIMERVALUE4(us_to_core_ticks(adap, s->timer_val[4])) |
		     TIMERVALUE5(us_to_core_ticks(adap, s->timer_val[5])));

	return 0;
}

int t4_sge_init(struct adapter *adap)
{
	struct sge *s = &adap->sge;
	u32 sge_control, sge_conm_ctrl;
	int ret, egress_threshold;

	/*
	 * Ingress Padding Boundary and Egress Status Page Size are set up by
	 * t4_fixup_host_params().
	 */
	sge_control = t4_read_reg(adap, SGE_CONTROL);
	s->pktshift = PKTSHIFT_GET(sge_control);
	s->stat_len = (sge_control & EGRSTATUSPAGESIZE_MASK) ? 128 : 64;
	s->fl_align = 1 << (INGPADBOUNDARY_GET(sge_control) +
			    X_INGPADBOUNDARY_SHIFT);

	if (adap->flags & USING_SOFT_PARAMS)
		ret = t4_sge_init_soft(adap);
	else
		ret = t4_sge_init_hard(adap);
	if (ret < 0)
		return ret;

	/*
	 * 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.) For T4,
	 * there was only a single field to control this.  For T5 there's the
	 * original field which now only applies to Unpacked Mode Free List
	 * buffers and a new field which only applies to Packed Mode Free List
	 * buffers.
	 */
	sge_conm_ctrl = t4_read_reg(adap, SGE_CONM_CTRL);
	if (is_t4(adap->params.chip))
		egress_threshold = EGRTHRESHOLD_GET(sge_conm_ctrl);
	else
		egress_threshold = EGRTHRESHOLDPACKING_GET(sge_conm_ctrl);
	s->fl_starve_thres = 2*egress_threshold + 1;

	setup_timer(&s->rx_timer, sge_rx_timer_cb, (unsigned long)adap);
	setup_timer(&s->tx_timer, sge_tx_timer_cb, (unsigned long)adap);
	s->idma_1s_thresh = core_ticks_per_usec(adap) * 1000000;  /* 1 s */
	s->idma_stalled[0] = 0;
	s->idma_stalled[1] = 0;
	spin_lock_init(&s->intrq_lock);

	return 0;
}