summaryrefslogtreecommitdiff
path: root/drivers/net/cxgb3/sge.c
blob: cf2e1d3c0d8d0abe96afaf0c99a1032cffc2e2ae (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
/*
 * Copyright (c) 2005-2008 Chelsio, 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/tcp.h>
#include <linux/dma-mapping.h>
#include <net/arp.h>
#include "common.h"
#include "regs.h"
#include "sge_defs.h"
#include "t3_cpl.h"
#include "firmware_exports.h"

#define USE_GTS 0

#define SGE_RX_SM_BUF_SIZE 1536

#define SGE_RX_COPY_THRES  256
#define SGE_RX_PULL_LEN    128

#define SGE_PG_RSVD SMP_CACHE_BYTES
/*
 * Page chunk size for FL0 buffers if FL0 is to be populated with page chunks.
 * It must be a divisor of PAGE_SIZE.  If set to 0 FL0 will use sk_buffs
 * directly.
 */
#define FL0_PG_CHUNK_SIZE  2048
#define FL0_PG_ORDER 0
#define FL0_PG_ALLOC_SIZE (PAGE_SIZE << FL0_PG_ORDER)
#define FL1_PG_CHUNK_SIZE (PAGE_SIZE > 8192 ? 16384 : 8192)
#define FL1_PG_ORDER (PAGE_SIZE > 8192 ? 0 : 1)
#define FL1_PG_ALLOC_SIZE (PAGE_SIZE << FL1_PG_ORDER)

#define SGE_RX_DROP_THRES 16
#define RX_RECLAIM_PERIOD (HZ/4)

/*
 * Max number of Rx buffers we replenish at a time.
 */
#define MAX_RX_REFILL 16U
/*
 * Period of the Tx buffer reclaim timer.  This timer does not need to run
 * frequently as Tx buffers are usually reclaimed by new Tx packets.
 */
#define TX_RECLAIM_PERIOD (HZ / 4)
#define TX_RECLAIM_TIMER_CHUNK 64U
#define TX_RECLAIM_CHUNK 16U

/* WR size in bytes */
#define WR_LEN (WR_FLITS * 8)

/*
 * Types of Tx queues in each queue set.  Order here matters, do not change.
 */
enum { TXQ_ETH, TXQ_OFLD, TXQ_CTRL };

/* Values for sge_txq.flags */
enum {
	TXQ_RUNNING = 1 << 0,	/* fetch engine is running */
	TXQ_LAST_PKT_DB = 1 << 1,	/* last packet rang the doorbell */
};

struct tx_desc {
	__be64 flit[TX_DESC_FLITS];
};

struct rx_desc {
	__be32 addr_lo;
	__be32 len_gen;
	__be32 gen2;
	__be32 addr_hi;
};

struct tx_sw_desc {		/* SW state per Tx descriptor */
	struct sk_buff *skb;
	u8 eop;       /* set if last descriptor for packet */
	u8 addr_idx;  /* buffer index of first SGL entry in descriptor */
	u8 fragidx;   /* first page fragment associated with descriptor */
	s8 sflit;     /* start flit of first SGL entry in descriptor */
};

struct rx_sw_desc {                /* SW state per Rx descriptor */
	union {
		struct sk_buff *skb;
		struct fl_pg_chunk pg_chunk;
	};
	DECLARE_PCI_UNMAP_ADDR(dma_addr);
};

struct rsp_desc {		/* response queue descriptor */
	struct rss_header rss_hdr;
	__be32 flags;
	__be32 len_cq;
	u8 imm_data[47];
	u8 intr_gen;
};

/*
 * Holds unmapping information for Tx packets that need deferred unmapping.
 * This structure lives at skb->head and must be allocated by callers.
 */
struct deferred_unmap_info {
	struct pci_dev *pdev;
	dma_addr_t addr[MAX_SKB_FRAGS + 1];
};

/*
 * Maps a number of flits to the number of Tx descriptors that can hold them.
 * The formula is
 *
 * desc = 1 + (flits - 2) / (WR_FLITS - 1).
 *
 * HW allows up to 4 descriptors to be combined into a WR.
 */
static u8 flit_desc_map[] = {
	0,
#if SGE_NUM_GENBITS == 1
	1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
	2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
	3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
	4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4
#elif SGE_NUM_GENBITS == 2
	1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
	2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
	3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
	4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
#else
# error "SGE_NUM_GENBITS must be 1 or 2"
#endif
};

static inline struct sge_qset *fl_to_qset(const struct sge_fl *q, int qidx)
{
	return container_of(q, struct sge_qset, fl[qidx]);
}

static inline struct sge_qset *rspq_to_qset(const struct sge_rspq *q)
{
	return container_of(q, struct sge_qset, rspq);
}

static inline struct sge_qset *txq_to_qset(const struct sge_txq *q, int qidx)
{
	return container_of(q, struct sge_qset, txq[qidx]);
}

/**
 *	refill_rspq - replenish an SGE response queue
 *	@adapter: the adapter
 *	@q: the response queue to replenish
 *	@credits: how many new responses to make available
 *
 *	Replenishes a response queue by making the supplied number of responses
 *	available to HW.
 */
static inline void refill_rspq(struct adapter *adapter,
			       const struct sge_rspq *q, unsigned int credits)
{
	rmb();
	t3_write_reg(adapter, A_SG_RSPQ_CREDIT_RETURN,
		     V_RSPQ(q->cntxt_id) | V_CREDITS(credits));
}

/**
 *	need_skb_unmap - does the platform need unmapping of sk_buffs?
 *
 *	Returns true if the platfrom needs sk_buff unmapping.  The compiler
 *	optimizes away unecessary code if this returns true.
 */
static inline int need_skb_unmap(void)
{
	/*
	 * This structure is used to tell if the platfrom needs buffer
	 * unmapping by checking if DECLARE_PCI_UNMAP_ADDR defines anything.
	 */
	struct dummy {
		DECLARE_PCI_UNMAP_ADDR(addr);
	};

	return sizeof(struct dummy) != 0;
}

/**
 *	unmap_skb - unmap a packet main body and its page fragments
 *	@skb: the packet
 *	@q: the Tx queue containing Tx descriptors for the packet
 *	@cidx: index of Tx descriptor
 *	@pdev: the PCI device
 *
 *	Unmap the main body of an sk_buff and its page fragments, if any.
 *	Because of the fairly complicated structure of our SGLs and the desire
 *	to conserve space for metadata, the information necessary to unmap an
 *	sk_buff is spread across the sk_buff itself (buffer lengths), the HW Tx
 *	descriptors (the physical addresses of the various data buffers), and
 *	the SW descriptor state (assorted indices).  The send functions
 *	initialize the indices for the first packet descriptor so we can unmap
 *	the buffers held in the first Tx descriptor here, and we have enough
 *	information at this point to set the state for the next Tx descriptor.
 *
 *	Note that it is possible to clean up the first descriptor of a packet
 *	before the send routines have written the next descriptors, but this
 *	race does not cause any problem.  We just end up writing the unmapping
 *	info for the descriptor first.
 */
static inline void unmap_skb(struct sk_buff *skb, struct sge_txq *q,
			     unsigned int cidx, struct pci_dev *pdev)
{
	const struct sg_ent *sgp;
	struct tx_sw_desc *d = &q->sdesc[cidx];
	int nfrags, frag_idx, curflit, j = d->addr_idx;

	sgp = (struct sg_ent *)&q->desc[cidx].flit[d->sflit];
	frag_idx = d->fragidx;

	if (frag_idx == 0 && skb_headlen(skb)) {
		pci_unmap_single(pdev, be64_to_cpu(sgp->addr[0]),
				 skb_headlen(skb), PCI_DMA_TODEVICE);
		j = 1;
	}

	curflit = d->sflit + 1 + j;
	nfrags = skb_shinfo(skb)->nr_frags;

	while (frag_idx < nfrags && curflit < WR_FLITS) {
		pci_unmap_page(pdev, be64_to_cpu(sgp->addr[j]),
			       skb_shinfo(skb)->frags[frag_idx].size,
			       PCI_DMA_TODEVICE);
		j ^= 1;
		if (j == 0) {
			sgp++;
			curflit++;
		}
		curflit++;
		frag_idx++;
	}

	if (frag_idx < nfrags) {   /* SGL continues into next Tx descriptor */
		d = cidx + 1 == q->size ? q->sdesc : d + 1;
		d->fragidx = frag_idx;
		d->addr_idx = j;
		d->sflit = curflit - WR_FLITS - j; /* sflit can be -1 */
	}
}

/**
 *	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
 *
 *	Reclaims Tx descriptors from an SGE Tx queue and frees the associated
 *	Tx buffers.  Called with the Tx queue lock held.
 */
static void free_tx_desc(struct adapter *adapter, struct sge_txq *q,
			 unsigned int n)
{
	struct tx_sw_desc *d;
	struct pci_dev *pdev = adapter->pdev;
	unsigned int cidx = q->cidx;

	const int need_unmap = need_skb_unmap() &&
			       q->cntxt_id >= FW_TUNNEL_SGEEC_START;

	d = &q->sdesc[cidx];
	while (n--) {
		if (d->skb) {	/* an SGL is present */
			if (need_unmap)
				unmap_skb(d->skb, q, cidx, pdev);
			if (d->eop)
				kfree_skb(d->skb);
		}
		++d;
		if (++cidx == q->size) {
			cidx = 0;
			d = q->sdesc;
		}
	}
	q->cidx = cidx;
}

/**
 *	reclaim_completed_tx - reclaims completed Tx descriptors
 *	@adapter: the adapter
 *	@q: the Tx queue to reclaim completed descriptors from
 *	@chunk: maximum number of descriptors to reclaim
 *
 *	Reclaims Tx descriptors that the SGE has indicated it has processed,
 *	and frees the associated buffers if possible.  Called with the Tx
 *	queue's lock held.
 */
static inline unsigned int reclaim_completed_tx(struct adapter *adapter,
						struct sge_txq *q,
						unsigned int chunk)
{
	unsigned int reclaim = q->processed - q->cleaned;

	reclaim = min(chunk, reclaim);
	if (reclaim) {
		free_tx_desc(adapter, q, reclaim);
		q->cleaned += reclaim;
		q->in_use -= reclaim;
	}
	return q->processed - q->cleaned;
}

/**
 *	should_restart_tx - are there enough resources to restart a Tx queue?
 *	@q: the Tx queue
 *
 *	Checks if there are enough descriptors to restart a suspended Tx queue.
 */
static inline int should_restart_tx(const struct sge_txq *q)
{
	unsigned int r = q->processed - q->cleaned;

	return q->in_use - r < (q->size >> 1);
}

static void clear_rx_desc(struct pci_dev *pdev, const struct sge_fl *q,
			  struct rx_sw_desc *d)
{
	if (q->use_pages && d->pg_chunk.page) {
		(*d->pg_chunk.p_cnt)--;
		if (!*d->pg_chunk.p_cnt)
			pci_unmap_page(pdev,
				       d->pg_chunk.mapping,
				       q->alloc_size, PCI_DMA_FROMDEVICE);

		put_page(d->pg_chunk.page);
		d->pg_chunk.page = NULL;
	} else {
		pci_unmap_single(pdev, pci_unmap_addr(d, dma_addr),
				 q->buf_size, PCI_DMA_FROMDEVICE);
		kfree_skb(d->skb);
		d->skb = NULL;
	}
}

/**
 *	free_rx_bufs - free the Rx buffers on an SGE free list
 *	@pdev: the PCI device associated with the adapter
 *	@rxq: the SGE free list to clean up
 *
 *	Release the buffers on an SGE free-buffer Rx queue.  HW fetching from
 *	this queue should be stopped before calling this function.
 */
static void free_rx_bufs(struct pci_dev *pdev, struct sge_fl *q)
{
	unsigned int cidx = q->cidx;

	while (q->credits--) {
		struct rx_sw_desc *d = &q->sdesc[cidx];


		clear_rx_desc(pdev, q, d);
		if (++cidx == q->size)
			cidx = 0;
	}

	if (q->pg_chunk.page) {
		__free_pages(q->pg_chunk.page, q->order);
		q->pg_chunk.page = NULL;
	}
}

/**
 *	add_one_rx_buf - add a packet buffer to a free-buffer list
 *	@va:  buffer start VA
 *	@len: the buffer length
 *	@d: the HW Rx descriptor to write
 *	@sd: the SW Rx descriptor to write
 *	@gen: the generation bit value
 *	@pdev: the PCI device associated with the adapter
 *
 *	Add a buffer of the given length to the supplied HW and SW Rx
 *	descriptors.
 */
static inline int add_one_rx_buf(void *va, unsigned int len,
				 struct rx_desc *d, struct rx_sw_desc *sd,
				 unsigned int gen, struct pci_dev *pdev)
{
	dma_addr_t mapping;

	mapping = pci_map_single(pdev, va, len, PCI_DMA_FROMDEVICE);
	if (unlikely(pci_dma_mapping_error(pdev, mapping)))
		return -ENOMEM;

	pci_unmap_addr_set(sd, dma_addr, mapping);

	d->addr_lo = cpu_to_be32(mapping);
	d->addr_hi = cpu_to_be32((u64) mapping >> 32);
	wmb();
	d->len_gen = cpu_to_be32(V_FLD_GEN1(gen));
	d->gen2 = cpu_to_be32(V_FLD_GEN2(gen));
	return 0;
}

static inline int add_one_rx_chunk(dma_addr_t mapping, struct rx_desc *d,
				   unsigned int gen)
{
	d->addr_lo = cpu_to_be32(mapping);
	d->addr_hi = cpu_to_be32((u64) mapping >> 32);
	wmb();
	d->len_gen = cpu_to_be32(V_FLD_GEN1(gen));
	d->gen2 = cpu_to_be32(V_FLD_GEN2(gen));
	return 0;
}

static int alloc_pg_chunk(struct adapter *adapter, struct sge_fl *q,
			  struct rx_sw_desc *sd, gfp_t gfp,
			  unsigned int order)
{
	if (!q->pg_chunk.page) {
		dma_addr_t mapping;

		q->pg_chunk.page = alloc_pages(gfp, order);
		if (unlikely(!q->pg_chunk.page))
			return -ENOMEM;
		q->pg_chunk.va = page_address(q->pg_chunk.page);
		q->pg_chunk.p_cnt = q->pg_chunk.va + (PAGE_SIZE << order) -
				    SGE_PG_RSVD;
		q->pg_chunk.offset = 0;
		mapping = pci_map_page(adapter->pdev, q->pg_chunk.page,
				       0, q->alloc_size, PCI_DMA_FROMDEVICE);
		q->pg_chunk.mapping = mapping;
	}
	sd->pg_chunk = q->pg_chunk;

	prefetch(sd->pg_chunk.p_cnt);

	q->pg_chunk.offset += q->buf_size;
	if (q->pg_chunk.offset == (PAGE_SIZE << order))
		q->pg_chunk.page = NULL;
	else {
		q->pg_chunk.va += q->buf_size;
		get_page(q->pg_chunk.page);
	}

	if (sd->pg_chunk.offset == 0)
		*sd->pg_chunk.p_cnt = 1;
	else
		*sd->pg_chunk.p_cnt += 1;

	return 0;
}

static inline void ring_fl_db(struct adapter *adap, struct sge_fl *q)
{
	if (q->pend_cred >= q->credits / 4) {
		q->pend_cred = 0;
		t3_write_reg(adap, A_SG_KDOORBELL, V_EGRCNTX(q->cntxt_id));
	}
}

/**
 *	refill_fl - refill an SGE free-buffer list
 *	@adapter: the adapter
 *	@q: the free-list to refill
 *	@n: the number of new buffers to allocate
 *	@gfp: the gfp flags for allocating new buffers
 *
 *	(Re)populate an SGE free-buffer list 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.
 */
static int refill_fl(struct adapter *adap, struct sge_fl *q, int n, gfp_t gfp)
{
	struct rx_sw_desc *sd = &q->sdesc[q->pidx];
	struct rx_desc *d = &q->desc[q->pidx];
	unsigned int count = 0;

	while (n--) {
		dma_addr_t mapping;
		int err;

		if (q->use_pages) {
			if (unlikely(alloc_pg_chunk(adap, q, sd, gfp,
						    q->order))) {
nomem:				q->alloc_failed++;
				break;
			}
			mapping = sd->pg_chunk.mapping + sd->pg_chunk.offset;
			pci_unmap_addr_set(sd, dma_addr, mapping);

			add_one_rx_chunk(mapping, d, q->gen);
			pci_dma_sync_single_for_device(adap->pdev, mapping,
						q->buf_size - SGE_PG_RSVD,
						PCI_DMA_FROMDEVICE);
		} else {
			void *buf_start;

			struct sk_buff *skb = alloc_skb(q->buf_size, gfp);
			if (!skb)
				goto nomem;

			sd->skb = skb;
			buf_start = skb->data;
			err = add_one_rx_buf(buf_start, q->buf_size, d, sd,
					     q->gen, adap->pdev);
			if (unlikely(err)) {
				clear_rx_desc(adap->pdev, q, sd);
				break;
			}
		}

		d++;
		sd++;
		if (++q->pidx == q->size) {
			q->pidx = 0;
			q->gen ^= 1;
			sd = q->sdesc;
			d = q->desc;
		}
		count++;
	}

	q->credits += count;
	q->pend_cred += count;
	ring_fl_db(adap, q);

	return count;
}

static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl)
{
	refill_fl(adap, fl, min(MAX_RX_REFILL, fl->size - fl->credits),
		  GFP_ATOMIC | __GFP_COMP);
}

/**
 *	recycle_rx_buf - recycle a receive buffer
 *	@adapter: the adapter
 *	@q: the SGE free list
 *	@idx: index of buffer to recycle
 *
 *	Recycles the specified buffer on the given free list by adding it at
 *	the next available slot on the list.
 */
static void recycle_rx_buf(struct adapter *adap, struct sge_fl *q,
			   unsigned int idx)
{
	struct rx_desc *from = &q->desc[idx];
	struct rx_desc *to = &q->desc[q->pidx];

	q->sdesc[q->pidx] = q->sdesc[idx];
	to->addr_lo = from->addr_lo;	/* already big endian */
	to->addr_hi = from->addr_hi;	/* likewise */
	wmb();
	to->len_gen = cpu_to_be32(V_FLD_GEN1(q->gen));
	to->gen2 = cpu_to_be32(V_FLD_GEN2(q->gen));

	if (++q->pidx == q->size) {
		q->pidx = 0;
		q->gen ^= 1;
	}

	q->credits++;
	q->pend_cred++;
	ring_fl_db(adap, q);
}

/**
 *	alloc_ring - allocate resources for an SGE descriptor ring
 *	@pdev: the PCI 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
 *
 *	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 physical address of the HW ring, and the address
 *	of the SW ring.
 */
static void *alloc_ring(struct pci_dev *pdev, size_t nelem, size_t elem_size,
			size_t sw_size, dma_addr_t * phys, void *metadata)
{
	size_t len = nelem * elem_size;
	void *s = NULL;
	void *p = dma_alloc_coherent(&pdev->dev, len, phys, GFP_KERNEL);

	if (!p)
		return NULL;
	if (sw_size && metadata) {
		s = kcalloc(nelem, sw_size, GFP_KERNEL);

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

/**
 *	t3_reset_qset - reset a sge qset
 *	@q: the queue set
 *
 *	Reset the qset structure.
 *	the NAPI structure is preserved in the event of
 *	the qset's reincarnation, for example during EEH recovery.
 */
static void t3_reset_qset(struct sge_qset *q)
{
	if (q->adap &&
	    !(q->adap->flags & NAPI_INIT)) {
		memset(q, 0, sizeof(*q));
		return;
	}

	q->adap = NULL;
	memset(&q->rspq, 0, sizeof(q->rspq));
	memset(q->fl, 0, sizeof(struct sge_fl) * SGE_RXQ_PER_SET);
	memset(q->txq, 0, sizeof(struct sge_txq) * SGE_TXQ_PER_SET);
	q->txq_stopped = 0;
	q->tx_reclaim_timer.function = NULL; /* for t3_stop_sge_timers() */
	q->rx_reclaim_timer.function = NULL;
	q->nomem = 0;
	napi_free_frags(&q->napi);
}


/**
 *	free_qset - free the resources of an SGE queue set
 *	@adapter: the adapter owning the queue set
 *	@q: the queue set
 *
 *	Release the HW and SW resources associated with an SGE queue set, such
 *	as HW contexts, packet buffers, and descriptor rings.  Traffic to the
 *	queue set must be quiesced prior to calling this.
 */
static void t3_free_qset(struct adapter *adapter, struct sge_qset *q)
{
	int i;
	struct pci_dev *pdev = adapter->pdev;

	for (i = 0; i < SGE_RXQ_PER_SET; ++i)
		if (q->fl[i].desc) {
			spin_lock_irq(&adapter->sge.reg_lock);
			t3_sge_disable_fl(adapter, q->fl[i].cntxt_id);
			spin_unlock_irq(&adapter->sge.reg_lock);
			free_rx_bufs(pdev, &q->fl[i]);
			kfree(q->fl[i].sdesc);
			dma_free_coherent(&pdev->dev,
					  q->fl[i].size *
					  sizeof(struct rx_desc), q->fl[i].desc,
					  q->fl[i].phys_addr);
		}

	for (i = 0; i < SGE_TXQ_PER_SET; ++i)
		if (q->txq[i].desc) {
			spin_lock_irq(&adapter->sge.reg_lock);
			t3_sge_enable_ecntxt(adapter, q->txq[i].cntxt_id, 0);
			spin_unlock_irq(&adapter->sge.reg_lock);
			if (q->txq[i].sdesc) {
				free_tx_desc(adapter, &q->txq[i],
					     q->txq[i].in_use);
				kfree(q->txq[i].sdesc);
			}
			dma_free_coherent(&pdev->dev,
					  q->txq[i].size *
					  sizeof(struct tx_desc),
					  q->txq[i].desc, q->txq[i].phys_addr);
			__skb_queue_purge(&q->txq[i].sendq);
		}

	if (q->rspq.desc) {
		spin_lock_irq(&adapter->sge.reg_lock);
		t3_sge_disable_rspcntxt(adapter, q->rspq.cntxt_id);
		spin_unlock_irq(&adapter->sge.reg_lock);
		dma_free_coherent(&pdev->dev,
				  q->rspq.size * sizeof(struct rsp_desc),
				  q->rspq.desc, q->rspq.phys_addr);
	}

	t3_reset_qset(q);
}

/**
 *	init_qset_cntxt - initialize an SGE queue set context info
 *	@qs: the queue set
 *	@id: the queue set id
 *
 *	Initializes the TIDs and context ids for the queues of a queue set.
 */
static void init_qset_cntxt(struct sge_qset *qs, unsigned int id)
{
	qs->rspq.cntxt_id = id;
	qs->fl[0].cntxt_id = 2 * id;
	qs->fl[1].cntxt_id = 2 * id + 1;
	qs->txq[TXQ_ETH].cntxt_id = FW_TUNNEL_SGEEC_START + id;
	qs->txq[TXQ_ETH].token = FW_TUNNEL_TID_START + id;
	qs->txq[TXQ_OFLD].cntxt_id = FW_OFLD_SGEEC_START + id;
	qs->txq[TXQ_CTRL].cntxt_id = FW_CTRL_SGEEC_START + id;
	qs->txq[TXQ_CTRL].token = FW_CTRL_TID_START + id;
}

/**
 *	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)
{
	/* alternatively: 3 * (n / 2) + 2 * (n & 1) */
	return (3 * n) / 2 + (n & 1);
}

/**
 *	flits_to_desc - returns the num of Tx descriptors for the given flits
 *	@n: the number of flits
 *
 *	Calculates 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 >= ARRAY_SIZE(flit_desc_map));
	return flit_desc_map[n];
}

/**
 *	get_packet - return the next ingress packet buffer from a free list
 *	@adap: the adapter that received the packet
 *	@fl: the SGE free list holding the packet
 *	@len: the packet length including any SGE padding
 *	@drop_thres: # of remaining buffers before we start dropping packets
 *
 *	Get the next packet from a free list and complete setup of the
 *	sk_buff.  If the packet is small we make a copy and recycle the
 *	original buffer, otherwise we use the original buffer itself.  If a
 *	positive drop threshold is supplied packets are dropped and their
 *	buffers recycled if (a) the number of remaining buffers is under the
 *	threshold and the packet is too big to copy, or (b) the packet should
 *	be copied but there is no memory for the copy.
 */
static struct sk_buff *get_packet(struct adapter *adap, struct sge_fl *fl,
				  unsigned int len, unsigned int drop_thres)
{
	struct sk_buff *skb = NULL;
	struct rx_sw_desc *sd = &fl->sdesc[fl->cidx];

	prefetch(sd->skb->data);
	fl->credits--;

	if (len <= SGE_RX_COPY_THRES) {
		skb = alloc_skb(len, GFP_ATOMIC);
		if (likely(skb != NULL)) {
			__skb_put(skb, len);
			pci_dma_sync_single_for_cpu(adap->pdev,
					    pci_unmap_addr(sd, dma_addr), len,
					    PCI_DMA_FROMDEVICE);
			memcpy(skb->data, sd->skb->data, len);
			pci_dma_sync_single_for_device(adap->pdev,
					    pci_unmap_addr(sd, dma_addr), len,
					    PCI_DMA_FROMDEVICE);
		} else if (!drop_thres)
			goto use_orig_buf;
recycle:
		recycle_rx_buf(adap, fl, fl->cidx);
		return skb;
	}

	if (unlikely(fl->credits < drop_thres) &&
	    refill_fl(adap, fl, min(MAX_RX_REFILL, fl->size - fl->credits - 1),
		      GFP_ATOMIC | __GFP_COMP) == 0)
		goto recycle;

use_orig_buf:
	pci_unmap_single(adap->pdev, pci_unmap_addr(sd, dma_addr),
			 fl->buf_size, PCI_DMA_FROMDEVICE);
	skb = sd->skb;
	skb_put(skb, len);
	__refill_fl(adap, fl);
	return skb;
}

/**
 *	get_packet_pg - return the next ingress packet buffer from a free list
 *	@adap: the adapter that received the packet
 *	@fl: the SGE free list holding the packet
 *	@len: the packet length including any SGE padding
 *	@drop_thres: # of remaining buffers before we start dropping packets
 *
 *	Get the next packet from a free list populated with page chunks.
 *	If the packet is small we make a copy and recycle the original buffer,
 *	otherwise we attach the original buffer as a page fragment to a fresh
 *	sk_buff.  If a positive drop threshold is supplied packets are dropped
 *	and their buffers recycled if (a) the number of remaining buffers is
 *	under the threshold and the packet is too big to copy, or (b) there's
 *	no system memory.
 *
 * 	Note: this function is similar to @get_packet but deals with Rx buffers
 * 	that are page chunks rather than sk_buffs.
 */
static struct sk_buff *get_packet_pg(struct adapter *adap, struct sge_fl *fl,
				     struct sge_rspq *q, unsigned int len,
				     unsigned int drop_thres)
{
	struct sk_buff *newskb, *skb;
	struct rx_sw_desc *sd = &fl->sdesc[fl->cidx];

	dma_addr_t dma_addr = pci_unmap_addr(sd, dma_addr);

	newskb = skb = q->pg_skb;
	if (!skb && (len <= SGE_RX_COPY_THRES)) {
		newskb = alloc_skb(len, GFP_ATOMIC);
		if (likely(newskb != NULL)) {
			__skb_put(newskb, len);
			pci_dma_sync_single_for_cpu(adap->pdev, dma_addr, len,
					    PCI_DMA_FROMDEVICE);
			memcpy(newskb->data, sd->pg_chunk.va, len);
			pci_dma_sync_single_for_device(adap->pdev, dma_addr,
						       len,
						       PCI_DMA_FROMDEVICE);
		} else if (!drop_thres)
			return NULL;
recycle:
		fl->credits--;
		recycle_rx_buf(adap, fl, fl->cidx);
		q->rx_recycle_buf++;
		return newskb;
	}

	if (unlikely(q->rx_recycle_buf || (!skb && fl->credits <= drop_thres)))
		goto recycle;

	prefetch(sd->pg_chunk.p_cnt);

	if (!skb)
		newskb = alloc_skb(SGE_RX_PULL_LEN, GFP_ATOMIC);

	if (unlikely(!newskb)) {
		if (!drop_thres)
			return NULL;
		goto recycle;
	}

	pci_dma_sync_single_for_cpu(adap->pdev, dma_addr, len,
				    PCI_DMA_FROMDEVICE);
	(*sd->pg_chunk.p_cnt)--;
	if (!*sd->pg_chunk.p_cnt)
		pci_unmap_page(adap->pdev,
			       sd->pg_chunk.mapping,
			       fl->alloc_size,
			       PCI_DMA_FROMDEVICE);
	if (!skb) {
		__skb_put(newskb, SGE_RX_PULL_LEN);
		memcpy(newskb->data, sd->pg_chunk.va, SGE_RX_PULL_LEN);
		skb_fill_page_desc(newskb, 0, sd->pg_chunk.page,
				   sd->pg_chunk.offset + SGE_RX_PULL_LEN,
				   len - SGE_RX_PULL_LEN);
		newskb->len = len;
		newskb->data_len = len - SGE_RX_PULL_LEN;
		newskb->truesize += newskb->data_len;
	} else {
		skb_fill_page_desc(newskb, skb_shinfo(newskb)->nr_frags,
				   sd->pg_chunk.page,
				   sd->pg_chunk.offset, len);
		newskb->len += len;
		newskb->data_len += len;
		newskb->truesize += len;
	}

	fl->credits--;
	/*
	 * We do not refill FLs here, we let the caller do it to overlap a
	 * prefetch.
	 */
	return newskb;
}

/**
 *	get_imm_packet - return the next ingress packet buffer from a response
 *	@resp: the response descriptor containing the packet data
 *
 *	Return a packet containing the immediate data of the given response.
 */
static inline struct sk_buff *get_imm_packet(const struct rsp_desc *resp)
{
	struct sk_buff *skb = alloc_skb(IMMED_PKT_SIZE, GFP_ATOMIC);

	if (skb) {
		__skb_put(skb, IMMED_PKT_SIZE);
		skb_copy_to_linear_data(skb, resp->imm_data, IMMED_PKT_SIZE);
	}
	return skb;
}

/**
 *	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.  Ethernet packets require addition of WR and CPL headers.
 */
static inline unsigned int calc_tx_descs(const struct sk_buff *skb)
{
	unsigned int flits;

	if (skb->len <= WR_LEN - sizeof(struct cpl_tx_pkt))
		return 1;

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

/**
 *	make_sgl - populate a scatter/gather list for a packet
 *	@skb: the packet
 *	@sgp: the SGL to populate
 *	@start: start address of skb main body data to include in the SGL
 *	@len: length of skb main body data to include in the SGL
 *	@pdev: the PCI device
 *
 *	Generates a scatter/gather list for the buffers that make up a packet
 *	and returns the SGL size in 8-byte words.  The caller must size the SGL
 *	appropriately.
 */
static inline unsigned int make_sgl(const struct sk_buff *skb,
				    struct sg_ent *sgp, unsigned char *start,
				    unsigned int len, struct pci_dev *pdev)
{
	dma_addr_t mapping;
	unsigned int i, j = 0, nfrags;

	if (len) {
		mapping = pci_map_single(pdev, start, len, PCI_DMA_TODEVICE);
		sgp->len[0] = cpu_to_be32(len);
		sgp->addr[0] = cpu_to_be64(mapping);
		j = 1;
	}

	nfrags = skb_shinfo(skb)->nr_frags;
	for (i = 0; i < nfrags; i++) {
		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];

		mapping = pci_map_page(pdev, frag->page, frag->page_offset,
				       frag->size, PCI_DMA_TODEVICE);
		sgp->len[j] = cpu_to_be32(frag->size);
		sgp->addr[j] = cpu_to_be64(mapping);
		j ^= 1;
		if (j == 0)
			++sgp;
	}
	if (j)
		sgp->len[j] = 0;
	return ((nfrags + (len != 0)) * 3) / 2 + j;
}

/**
 *	check_ring_tx_db - check and potentially ring a Tx queue's doorbell
 *	@adap: the adapter
 *	@q: the Tx queue
 *
 *	Ring the doorbel if a Tx queue is asleep.  There is a natural race,
 *	where the HW is going to sleep just after we checked, however,
 *	then the interrupt handler will detect the outstanding TX packet
 *	and ring the doorbell for us.
 *
 *	When GTS is disabled we unconditionally ring the doorbell.
 */
static inline void check_ring_tx_db(struct adapter *adap, struct sge_txq *q)
{
#if USE_GTS
	clear_bit(TXQ_LAST_PKT_DB, &q->flags);
	if (test_and_set_bit(TXQ_RUNNING, &q->flags) == 0) {
		set_bit(TXQ_LAST_PKT_DB, &q->flags);
		t3_write_reg(adap, A_SG_KDOORBELL,
			     F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id));
	}
#else
	wmb();			/* write descriptors before telling HW */
	t3_write_reg(adap, A_SG_KDOORBELL,
		     F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id));
#endif
}

static inline void wr_gen2(struct tx_desc *d, unsigned int gen)
{
#if SGE_NUM_GENBITS == 2
	d->flit[TX_DESC_FLITS - 1] = cpu_to_be64(gen);
#endif
}

/**
 *	write_wr_hdr_sgl - write a WR header and, optionally, SGL
 *	@ndesc: number of Tx descriptors spanned by the SGL
 *	@skb: the packet corresponding to the WR
 *	@d: first Tx descriptor to be written
 *	@pidx: index of above descriptors
 *	@q: the SGE Tx queue
 *	@sgl: the SGL
 *	@flits: number of flits to the start of the SGL in the first descriptor
 *	@sgl_flits: the SGL size in flits
 *	@gen: the Tx descriptor generation
 *	@wr_hi: top 32 bits of WR header based on WR type (big endian)
 *	@wr_lo: low 32 bits of WR header based on WR type (big endian)
 *
 *	Write a work request header and an associated SGL.  If the SGL is
 *	small enough to fit into one Tx descriptor it has already been written
 *	and we just need to write the WR header.  Otherwise we distribute the
 *	SGL across the number of descriptors it spans.
 */
static void write_wr_hdr_sgl(unsigned int ndesc, struct sk_buff *skb,
			     struct tx_desc *d, unsigned int pidx,
			     const struct sge_txq *q,
			     const struct sg_ent *sgl,
			     unsigned int flits, unsigned int sgl_flits,
			     unsigned int gen, __be32 wr_hi,
			     __be32 wr_lo)
{
	struct work_request_hdr *wrp = (struct work_request_hdr *)d;
	struct tx_sw_desc *sd = &q->sdesc[pidx];

	sd->skb = skb;
	if (need_skb_unmap()) {
		sd->fragidx = 0;
		sd->addr_idx = 0;
		sd->sflit = flits;
	}

	if (likely(ndesc == 1)) {
		sd->eop = 1;
		wrp->wr_hi = htonl(F_WR_SOP | F_WR_EOP | V_WR_DATATYPE(1) |
				   V_WR_SGLSFLT(flits)) | wr_hi;
		wmb();
		wrp->wr_lo = htonl(V_WR_LEN(flits + sgl_flits) |
				   V_WR_GEN(gen)) | wr_lo;
		wr_gen2(d, gen);
	} else {
		unsigned int ogen = gen;
		const u64 *fp = (const u64 *)sgl;
		struct work_request_hdr *wp = wrp;

		wrp->wr_hi = htonl(F_WR_SOP | V_WR_DATATYPE(1) |
				   V_WR_SGLSFLT(flits)) | wr_hi;

		while (sgl_flits) {
			unsigned int avail = WR_FLITS - flits;

			if (avail > sgl_flits)
				avail = sgl_flits;
			memcpy(&d->flit[flits], fp, avail * sizeof(*fp));
			sgl_flits -= avail;
			ndesc--;
			if (!sgl_flits)
				break;

			fp += avail;
			d++;
			sd->eop = 0;
			sd++;
			if (++pidx == q->size) {
				pidx = 0;
				gen ^= 1;
				d = q->desc;
				sd = q->sdesc;
			}

			sd->skb = skb;
			wrp = (struct work_request_hdr *)d;
			wrp->wr_hi = htonl(V_WR_DATATYPE(1) |
					   V_WR_SGLSFLT(1)) | wr_hi;
			wrp->wr_lo = htonl(V_WR_LEN(min(WR_FLITS,
							sgl_flits + 1)) |
					   V_WR_GEN(gen)) | wr_lo;
			wr_gen2(d, gen);
			flits = 1;
		}
		sd->eop = 1;
		wrp->wr_hi |= htonl(F_WR_EOP);
		wmb();
		wp->wr_lo = htonl(V_WR_LEN(WR_FLITS) | V_WR_GEN(ogen)) | wr_lo;
		wr_gen2((struct tx_desc *)wp, ogen);
		WARN_ON(ndesc != 0);
	}
}

/**
 *	write_tx_pkt_wr - write a TX_PKT work request
 *	@adap: the adapter
 *	@skb: the packet to send
 *	@pi: the egress interface
 *	@pidx: index of the first Tx descriptor to write
 *	@gen: the generation value to use
 *	@q: the Tx queue
 *	@ndesc: number of descriptors the packet will occupy
 *	@compl: the value of the COMPL bit to use
 *
 *	Generate a TX_PKT work request to send the supplied packet.
 */
static void write_tx_pkt_wr(struct adapter *adap, struct sk_buff *skb,
			    const struct port_info *pi,
			    unsigned int pidx, unsigned int gen,
			    struct sge_txq *q, unsigned int ndesc,
			    unsigned int compl)
{
	unsigned int flits, sgl_flits, cntrl, tso_info;
	struct sg_ent *sgp, sgl[MAX_SKB_FRAGS / 2 + 1];
	struct tx_desc *d = &q->desc[pidx];
	struct cpl_tx_pkt *cpl = (struct cpl_tx_pkt *)d;

	cpl->len = htonl(skb->len);
	cntrl = V_TXPKT_INTF(pi->port_id);

	if (vlan_tx_tag_present(skb) && pi->vlan_grp)
		cntrl |= F_TXPKT_VLAN_VLD | V_TXPKT_VLAN(vlan_tx_tag_get(skb));

	tso_info = V_LSO_MSS(skb_shinfo(skb)->gso_size);
	if (tso_info) {
		int eth_type;
		struct cpl_tx_pkt_lso *hdr = (struct cpl_tx_pkt_lso *)cpl;

		d->flit[2] = 0;
		cntrl |= V_TXPKT_OPCODE(CPL_TX_PKT_LSO);
		hdr->cntrl = htonl(cntrl);
		eth_type = skb_network_offset(skb) == ETH_HLEN ?
		    CPL_ETH_II : CPL_ETH_II_VLAN;
		tso_info |= V_LSO_ETH_TYPE(eth_type) |
		    V_LSO_IPHDR_WORDS(ip_hdr(skb)->ihl) |
		    V_LSO_TCPHDR_WORDS(tcp_hdr(skb)->doff);
		hdr->lso_info = htonl(tso_info);
		flits = 3;
	} else {
		cntrl |= V_TXPKT_OPCODE(CPL_TX_PKT);
		cntrl |= F_TXPKT_IPCSUM_DIS;	/* SW calculates IP csum */
		cntrl |= V_TXPKT_L4CSUM_DIS(skb->ip_summed != CHECKSUM_PARTIAL);
		cpl->cntrl = htonl(cntrl);

		if (skb->len <= WR_LEN - sizeof(*cpl)) {
			q->sdesc[pidx].skb = NULL;
			if (!skb->data_len)
				skb_copy_from_linear_data(skb, &d->flit[2],
							  skb->len);
			else
				skb_copy_bits(skb, 0, &d->flit[2], skb->len);

			flits = (skb->len + 7) / 8 + 2;
			cpl->wr.wr_hi = htonl(V_WR_BCNTLFLT(skb->len & 7) |
					      V_WR_OP(FW_WROPCODE_TUNNEL_TX_PKT)
					      | F_WR_SOP | F_WR_EOP | compl);
			wmb();
			cpl->wr.wr_lo = htonl(V_WR_LEN(flits) | V_WR_GEN(gen) |
					      V_WR_TID(q->token));
			wr_gen2(d, gen);
			kfree_skb(skb);
			return;
		}

		flits = 2;
	}

	sgp = ndesc == 1 ? (struct sg_ent *)&d->flit[flits] : sgl;
	sgl_flits = make_sgl(skb, sgp, skb->data, skb_headlen(skb), adap->pdev);

	write_wr_hdr_sgl(ndesc, skb, d, pidx, q, sgl, flits, sgl_flits, gen,
			 htonl(V_WR_OP(FW_WROPCODE_TUNNEL_TX_PKT) | compl),
			 htonl(V_WR_TID(q->token)));
}

static inline void t3_stop_tx_queue(struct netdev_queue *txq,
				    struct sge_qset *qs, struct sge_txq *q)
{
	netif_tx_stop_queue(txq);
	set_bit(TXQ_ETH, &qs->txq_stopped);
	q->stops++;
}

/**
 *	eth_xmit - add a packet to the Ethernet Tx queue
 *	@skb: the packet
 *	@dev: the egress net device
 *
 *	Add a packet to an SGE Tx queue.  Runs with softirqs disabled.
 */
netdev_tx_t t3_eth_xmit(struct sk_buff *skb, struct net_device *dev)
{
	int qidx;
	unsigned int ndesc, pidx, credits, gen, compl;
	const struct port_info *pi = netdev_priv(dev);
	struct adapter *adap = pi->adapter;
	struct netdev_queue *txq;
	struct sge_qset *qs;
	struct sge_txq *q;

	/*
	 * The chip min packet length is 9 octets but play safe and reject
	 * anything shorter than an Ethernet header.
	 */
	if (unlikely(skb->len < ETH_HLEN)) {
		dev_kfree_skb(skb);
		return NETDEV_TX_OK;
	}

	qidx = skb_get_queue_mapping(skb);
	qs = &pi->qs[qidx];
	q = &qs->txq[TXQ_ETH];
	txq = netdev_get_tx_queue(dev, qidx);

	reclaim_completed_tx(adap, q, TX_RECLAIM_CHUNK);

	credits = q->size - q->in_use;
	ndesc = calc_tx_descs(skb);

	if (unlikely(credits < ndesc)) {
		t3_stop_tx_queue(txq, qs, q);
		dev_err(&adap->pdev->dev,
			"%s: Tx ring %u full while queue awake!\n",
			dev->name, q->cntxt_id & 7);
		return NETDEV_TX_BUSY;
	}

	q->in_use += ndesc;
	if (unlikely(credits - ndesc < q->stop_thres)) {
		t3_stop_tx_queue(txq, qs, q);

		if (should_restart_tx(q) &&
		    test_and_clear_bit(TXQ_ETH, &qs->txq_stopped)) {
			q->restarts++;
			netif_tx_start_queue(txq);
		}
	}

	gen = q->gen;
	q->unacked += ndesc;
	compl = (q->unacked & 8) << (S_WR_COMPL - 3);
	q->unacked &= 7;
	pidx = q->pidx;
	q->pidx += ndesc;
	if (q->pidx >= q->size) {
		q->pidx -= q->size;
		q->gen ^= 1;
	}

	/* update port statistics */
	if (skb->ip_summed == CHECKSUM_COMPLETE)
		qs->port_stats[SGE_PSTAT_TX_CSUM]++;
	if (skb_shinfo(skb)->gso_size)
		qs->port_stats[SGE_PSTAT_TSO]++;
	if (vlan_tx_tag_present(skb) && pi->vlan_grp)
		qs->port_stats[SGE_PSTAT_VLANINS]++;

	/*
	 * We do not use Tx completion interrupts to free DMAd Tx packets.
	 * This is good for performamce but means that we rely on new Tx
	 * packets arriving to run the destructors of completed packets,
	 * which open up space in their sockets' send queues.  Sometimes
	 * we do not get such new packets causing Tx to stall.  A single
	 * UDP transmitter is a good example of this situation.  We have
	 * a clean up timer that periodically reclaims completed packets
	 * but it doesn't run often enough (nor do we want it to) to prevent
	 * lengthy stalls.  A solution to this problem is to run the
	 * destructor early, after the packet is queued but before it's DMAd.
	 * A cons is that we lie to socket memory accounting, but the amount
	 * of extra memory is reasonable (limited by the number of Tx
	 * descriptors), the packets do actually get freed quickly by new
	 * packets almost always, and for protocols like TCP that wait for
	 * acks to really free up the data the extra memory is even less.
	 * On the positive side we run the destructors on the sending CPU
	 * rather than on a potentially different completing CPU, usually a
	 * good thing.  We also run them without holding our Tx queue lock,
	 * unlike what reclaim_completed_tx() would otherwise do.
	 *
	 * Run the destructor before telling the DMA engine about the packet
	 * to make sure it doesn't complete and get freed prematurely.
	 */
	if (likely(!skb_shared(skb)))
		skb_orphan(skb);

	write_tx_pkt_wr(adap, skb, pi, pidx, gen, q, ndesc, compl);
	check_ring_tx_db(adap, q);
	return NETDEV_TX_OK;
}

/**
 *	write_imm - write a packet into a Tx descriptor as immediate data
 *	@d: the Tx descriptor to write
 *	@skb: the packet
 *	@len: the length of packet data to write as immediate data
 *	@gen: the generation bit value to write
 *
 *	Writes a packet as immediate data into a Tx descriptor.  The packet
 *	contains a work request at its beginning.  We must write the packet
 *	carefully so the SGE doesn't read it accidentally before it's written
 *	in its entirety.
 */
static inline void write_imm(struct tx_desc *d, struct sk_buff *skb,
			     unsigned int len, unsigned int gen)
{
	struct work_request_hdr *from = (struct work_request_hdr *)skb->data;
	struct work_request_hdr *to = (struct work_request_hdr *)d;

	if (likely(!skb->data_len))
		memcpy(&to[1], &from[1], len - sizeof(*from));
	else
		skb_copy_bits(skb, sizeof(*from), &to[1], len - sizeof(*from));

	to->wr_hi = from->wr_hi | htonl(F_WR_SOP | F_WR_EOP |
					V_WR_BCNTLFLT(len & 7));
	wmb();
	to->wr_lo = from->wr_lo | htonl(V_WR_GEN(gen) |
					V_WR_LEN((len + 7) / 8));
	wr_gen2(d, gen);
	kfree_skb(skb);
}

/**
 *	check_desc_avail - check descriptor availability on a send queue
 *	@adap: the adapter
 *	@q: the send queue
 *	@skb: the packet needing the descriptors
 *	@ndesc: the number of Tx descriptors needed
 *	@qid: the Tx queue number in its queue set (TXQ_OFLD or TXQ_CTRL)
 *
 *	Checks if the requested number of Tx descriptors is available on an
 *	SGE send queue.  If the queue is already suspended or not enough
 *	descriptors are available the packet is queued for later transmission.
 *	Must be called with the Tx queue locked.
 *
 *	Returns 0 if enough descriptors are available, 1 if there aren't
 *	enough descriptors and the packet has been queued, and 2 if the caller
 *	needs to retry because there weren't enough descriptors at the
 *	beginning of the call but some freed up in the mean time.
 */
static inline int check_desc_avail(struct adapter *adap, struct sge_txq *q,
				   struct sk_buff *skb, unsigned int ndesc,
				   unsigned int qid)
{
	if (unlikely(!skb_queue_empty(&q->sendq))) {
	      addq_exit:__skb_queue_tail(&q->sendq, skb);
		return 1;
	}
	if (unlikely(q->size - q->in_use < ndesc)) {
		struct sge_qset *qs = txq_to_qset(q, qid);

		set_bit(qid, &qs->txq_stopped);
		smp_mb__after_clear_bit();

		if (should_restart_tx(q) &&
		    test_and_clear_bit(qid, &qs->txq_stopped))
			return 2;

		q->stops++;
		goto addq_exit;
	}
	return 0;
}

/**
 *	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)
{
	unsigned int reclaim = q->processed - q->cleaned;

	q->in_use -= reclaim;
	q->cleaned += reclaim;
}

static inline int immediate(const struct sk_buff *skb)
{
	return skb->len <= WR_LEN;
}

/**
 *	ctrl_xmit - send a packet through an SGE control Tx queue
 *	@adap: the adapter
 *	@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 in a single Tx
 *	descriptor and have no page fragments.
 */
static int ctrl_xmit(struct adapter *adap, struct sge_txq *q,
		     struct sk_buff *skb)
{
	int ret;
	struct work_request_hdr *wrp = (struct work_request_hdr *)skb->data;

	if (unlikely(!immediate(skb))) {
		WARN_ON(1);
		dev_kfree_skb(skb);
		return NET_XMIT_SUCCESS;
	}

	wrp->wr_hi |= htonl(F_WR_SOP | F_WR_EOP);
	wrp->wr_lo = htonl(V_WR_TID(q->token));

	spin_lock(&q->lock);
      again:reclaim_completed_tx_imm(q);

	ret = check_desc_avail(adap, q, skb, 1, TXQ_CTRL);
	if (unlikely(ret)) {
		if (ret == 1) {
			spin_unlock(&q->lock);
			return NET_XMIT_CN;
		}
		goto again;
	}

	write_imm(&q->desc[q->pidx], skb, skb->len, q->gen);

	q->in_use++;
	if (++q->pidx >= q->size) {
		q->pidx = 0;
		q->gen ^= 1;
	}
	spin_unlock(&q->lock);
	wmb();
	t3_write_reg(adap, A_SG_KDOORBELL,
		     F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id));
	return NET_XMIT_SUCCESS;
}

/**
 *	restart_ctrlq - restart a suspended control queue
 *	@qs: the queue set cotaining the control queue
 *
 *	Resumes transmission on a suspended Tx control queue.
 */
static void restart_ctrlq(unsigned long data)
{
	struct sk_buff *skb;
	struct sge_qset *qs = (struct sge_qset *)data;
	struct sge_txq *q = &qs->txq[TXQ_CTRL];

	spin_lock(&q->lock);
      again:reclaim_completed_tx_imm(q);

	while (q->in_use < q->size &&
	       (skb = __skb_dequeue(&q->sendq)) != NULL) {

		write_imm(&q->desc[q->pidx], skb, skb->len, q->gen);

		if (++q->pidx >= q->size) {
			q->pidx = 0;
			q->gen ^= 1;
		}
		q->in_use++;
	}

	if (!skb_queue_empty(&q->sendq)) {
		set_bit(TXQ_CTRL, &qs->txq_stopped);
		smp_mb__after_clear_bit();

		if (should_restart_tx(q) &&
		    test_and_clear_bit(TXQ_CTRL, &qs->txq_stopped))
			goto again;
		q->stops++;
	}

	spin_unlock(&q->lock);
	wmb();
	t3_write_reg(qs->adap, A_SG_KDOORBELL,
		     F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id));
}

/*
 * Send a management message through control queue 0
 */
int t3_mgmt_tx(struct adapter *adap, struct sk_buff *skb)
{
	int ret;
	local_bh_disable();
	ret = ctrl_xmit(adap, &adap->sge.qs[0].txq[TXQ_CTRL], skb);
	local_bh_enable();

	return ret;
}

/**
 *	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)
{
	int i;
	const dma_addr_t *p;
	const struct skb_shared_info *si;
	const struct deferred_unmap_info *dui;

	dui = (struct deferred_unmap_info *)skb->head;
	p = dui->addr;

	if (skb->tail - skb->transport_header)
		pci_unmap_single(dui->pdev, *p++,
				 skb->tail - skb->transport_header,
				 PCI_DMA_TODEVICE);

	si = skb_shinfo(skb);
	for (i = 0; i < si->nr_frags; i++)
		pci_unmap_page(dui->pdev, *p++, si->frags[i].size,
			       PCI_DMA_TODEVICE);
}

static void setup_deferred_unmapping(struct sk_buff *skb, struct pci_dev *pdev,
				     const struct sg_ent *sgl, int sgl_flits)
{
	dma_addr_t *p;
	struct deferred_unmap_info *dui;

	dui = (struct deferred_unmap_info *)skb->head;
	dui->pdev = pdev;
	for (p = dui->addr; sgl_flits >= 3; sgl++, sgl_flits -= 3) {
		*p++ = be64_to_cpu(sgl->addr[0]);
		*p++ = be64_to_cpu(sgl->addr[1]);
	}
	if (sgl_flits)
		*p = be64_to_cpu(sgl->addr[0]);
}

/**
 *	write_ofld_wr - write an offload work request
 *	@adap: the adapter
 *	@skb: the packet to send
 *	@q: the Tx queue
 *	@pidx: index of the first Tx descriptor to write
 *	@gen: the generation value to use
 *	@ndesc: number of descriptors the packet will occupy
 *
 *	Write an offload work request to send the supplied packet.  The packet
 *	data already carry the work request with most fields populated.
 */
static void write_ofld_wr(struct adapter *adap, struct sk_buff *skb,
			  struct sge_txq *q, unsigned int pidx,
			  unsigned int gen, unsigned int ndesc)
{
	unsigned int sgl_flits, flits;
	struct work_request_hdr *from;
	struct sg_ent *sgp, sgl[MAX_SKB_FRAGS / 2 + 1];
	struct tx_desc *d = &q->desc[pidx];

	if (immediate(skb)) {
		q->sdesc[pidx].skb = NULL;
		write_imm(d, skb, skb->len, gen);
		return;
	}

	/* Only TX_DATA builds SGLs */

	from = (struct work_request_hdr *)skb->data;
	memcpy(&d->flit[1], &from[1],
	       skb_transport_offset(skb) - sizeof(*from));

	flits = skb_transport_offset(skb) / 8;
	sgp = ndesc == 1 ? (struct sg_ent *)&d->flit[flits] : sgl;
	sgl_flits = make_sgl(skb, sgp, skb_transport_header(skb),
			     skb->tail - skb->transport_header,
			     adap->pdev);
	if (need_skb_unmap()) {
		setup_deferred_unmapping(skb, adap->pdev, sgp, sgl_flits);
		skb->destructor = deferred_unmap_destructor;
	}

	write_wr_hdr_sgl(ndesc, skb, d, pidx, q, sgl, flits, sgl_flits,
			 gen, from->wr_hi, from->wr_lo);
}

/**
 *	calc_tx_descs_ofld - calculate # of Tx descriptors for an offload packet
 *	@skb: the packet
 *
 * 	Returns the number of Tx descriptors needed for the given offload
 * 	packet.  These packets are already fully constructed.
 */
static inline unsigned int calc_tx_descs_ofld(const struct sk_buff *skb)
{
	unsigned int flits, cnt;

	if (skb->len <= WR_LEN)
		return 1;	/* packet fits as immediate data */

	flits = skb_transport_offset(skb) / 8;	/* headers */
	cnt = skb_shinfo(skb)->nr_frags;
	if (skb->tail != skb->transport_header)
		cnt++;
	return flits_to_desc(flits + sgl_len(cnt));
}

/**
 *	ofld_xmit - send a packet through an offload queue
 *	@adap: the adapter
 *	@q: the Tx offload queue
 *	@skb: the packet
 *
 *	Send an offload packet through an SGE offload queue.
 */
static int ofld_xmit(struct adapter *adap, struct sge_txq *q,
		     struct sk_buff *skb)
{
	int ret;
	unsigned int ndesc = calc_tx_descs_ofld(skb), pidx, gen;

	spin_lock(&q->lock);
again:	reclaim_completed_tx(adap, q, TX_RECLAIM_CHUNK);

	ret = check_desc_avail(adap, q, skb, ndesc, TXQ_OFLD);
	if (unlikely(ret)) {
		if (ret == 1) {
			skb->priority = ndesc;	/* save for restart */
			spin_unlock(&q->lock);
			return NET_XMIT_CN;
		}
		goto again;
	}

	gen = q->gen;
	q->in_use += ndesc;
	pidx = q->pidx;
	q->pidx += ndesc;
	if (q->pidx >= q->size) {
		q->pidx -= q->size;
		q->gen ^= 1;
	}
	spin_unlock(&q->lock);

	write_ofld_wr(adap, skb, q, pidx, gen, ndesc);
	check_ring_tx_db(adap, q);
	return NET_XMIT_SUCCESS;
}

/**
 *	restart_offloadq - restart a suspended offload queue
 *	@qs: the queue set cotaining the offload queue
 *
 *	Resumes transmission on a suspended Tx offload queue.
 */
static void restart_offloadq(unsigned long data)
{
	struct sk_buff *skb;
	struct sge_qset *qs = (struct sge_qset *)data;
	struct sge_txq *q = &qs->txq[TXQ_OFLD];
	const struct port_info *pi = netdev_priv(qs->netdev);
	struct adapter *adap = pi->adapter;

	spin_lock(&q->lock);
again:	reclaim_completed_tx(adap, q, TX_RECLAIM_CHUNK);

	while ((skb = skb_peek(&q->sendq)) != NULL) {
		unsigned int gen, pidx;
		unsigned int ndesc = skb->priority;

		if (unlikely(q->size - q->in_use < ndesc)) {
			set_bit(TXQ_OFLD, &qs->txq_stopped);
			smp_mb__after_clear_bit();

			if (should_restart_tx(q) &&
			    test_and_clear_bit(TXQ_OFLD, &qs->txq_stopped))
				goto again;
			q->stops++;
			break;
		}

		gen = q->gen;
		q->in_use += ndesc;
		pidx = q->pidx;
		q->pidx += ndesc;
		if (q->pidx >= q->size) {
			q->pidx -= q->size;
			q->gen ^= 1;
		}
		__skb_unlink(skb, &q->sendq);
		spin_unlock(&q->lock);

		write_ofld_wr(adap, skb, q, pidx, gen, ndesc);
		spin_lock(&q->lock);
	}
	spin_unlock(&q->lock);

#if USE_GTS
	set_bit(TXQ_RUNNING, &q->flags);
	set_bit(TXQ_LAST_PKT_DB, &q->flags);
#endif
	wmb();
	t3_write_reg(adap, A_SG_KDOORBELL,
		     F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id));
}

/**
 *	queue_set - return the queue set a packet should use
 *	@skb: the packet
 *
 *	Maps a packet to the SGE queue set it should use.  The desired queue
 *	set is carried in bits 1-3 in the packet's priority.
 */
static inline int queue_set(const struct sk_buff *skb)
{
	return skb->priority >> 1;
}

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

/**
 *	t3_offload_tx - send an offload packet
 *	@tdev: the offload device to send to
 *	@skb: the packet
 *
 *	Sends an offload packet.  We use the packet priority to select the
 *	appropriate Tx queue as follows: bit 0 indicates whether the packet
 *	should be sent as regular or control, bits 1-3 select the queue set.
 */
int t3_offload_tx(struct t3cdev *tdev, struct sk_buff *skb)
{
	struct adapter *adap = tdev2adap(tdev);
	struct sge_qset *qs = &adap->sge.qs[queue_set(skb)];

	if (unlikely(is_ctrl_pkt(skb)))
		return ctrl_xmit(adap, &qs->txq[TXQ_CTRL], skb);

	return ofld_xmit(adap, &qs->txq[TXQ_OFLD], skb);
}

/**
 *	offload_enqueue - add an offload packet to an SGE offload receive queue
 *	@q: the SGE response queue
 *	@skb: the packet
 *
 *	Add a new offload packet to an SGE response queue's offload packet
 *	queue.  If the packet is the first on the queue it schedules the RX
 *	softirq to process the queue.
 */
static inline void offload_enqueue(struct sge_rspq *q, struct sk_buff *skb)
{
	int was_empty = skb_queue_empty(&q->rx_queue);

	__skb_queue_tail(&q->rx_queue, skb);

	if (was_empty) {
		struct sge_qset *qs = rspq_to_qset(q);

		napi_schedule(&qs->napi);
	}
}

/**
 *	deliver_partial_bundle - deliver a (partial) bundle of Rx offload pkts
 *	@tdev: the offload device that will be receiving the packets
 *	@q: the SGE response queue that assembled the bundle
 *	@skbs: the partial bundle
 *	@n: the number of packets in the bundle
 *
 *	Delivers a (partial) bundle of Rx offload packets to an offload device.
 */
static inline void deliver_partial_bundle(struct t3cdev *tdev,
					  struct sge_rspq *q,
					  struct sk_buff *skbs[], int n)
{
	if (n) {
		q->offload_bundles++;
		tdev->recv(tdev, skbs, n);
	}
}

/**
 *	ofld_poll - NAPI handler for offload packets in interrupt mode
 *	@dev: the network device doing the polling
 *	@budget: polling budget
 *
 *	The NAPI handler for offload packets when a response queue is serviced
 *	by the hard interrupt handler, i.e., when it's operating in non-polling
 *	mode.  Creates small packet batches and sends them through the offload
 *	receive handler.  Batches need to be of modest size as we do prefetches
 *	on the packets in each.
 */
static int ofld_poll(struct napi_struct *napi, int budget)
{
	struct sge_qset *qs = container_of(napi, struct sge_qset, napi);
	struct sge_rspq *q = &qs->rspq;
	struct adapter *adapter = qs->adap;
	int work_done = 0;

	while (work_done < budget) {
		struct sk_buff *skb, *tmp, *skbs[RX_BUNDLE_SIZE];
		struct sk_buff_head queue;
		int ngathered;

		spin_lock_irq(&q->lock);
		__skb_queue_head_init(&queue);
		skb_queue_splice_init(&q->rx_queue, &queue);
		if (skb_queue_empty(&queue)) {
			napi_complete(napi);
			spin_unlock_irq(&q->lock);
			return work_done;
		}
		spin_unlock_irq(&q->lock);

		ngathered = 0;
		skb_queue_walk_safe(&queue, skb, tmp) {
			if (work_done >= budget)
				break;
			work_done++;

			__skb_unlink(skb, &queue);
			prefetch(skb->data);
			skbs[ngathered] = skb;
			if (++ngathered == RX_BUNDLE_SIZE) {
				q->offload_bundles++;
				adapter->tdev.recv(&adapter->tdev, skbs,
						   ngathered);
				ngathered = 0;
			}
		}
		if (!skb_queue_empty(&queue)) {
			/* splice remaining packets back onto Rx queue */
			spin_lock_irq(&q->lock);
			skb_queue_splice(&queue, &q->rx_queue);
			spin_unlock_irq(&q->lock);
		}
		deliver_partial_bundle(&adapter->tdev, q, skbs, ngathered);
	}

	return work_done;
}

/**
 *	rx_offload - process a received offload packet
 *	@tdev: the offload device receiving the packet
 *	@rq: the response queue that received the packet
 *	@skb: the packet
 *	@rx_gather: a gather list of packets if we are building a bundle
 *	@gather_idx: index of the next available slot in the bundle
 *
 *	Process an ingress offload pakcet and add it to the offload ingress
 *	queue. 	Returns the index of the next available slot in the bundle.
 */
static inline int rx_offload(struct t3cdev *tdev, struct sge_rspq *rq,
			     struct sk_buff *skb, struct sk_buff *rx_gather[],
			     unsigned int gather_idx)
{
	skb_reset_mac_header(skb);
	skb_reset_network_header(skb);
	skb_reset_transport_header(skb);

	if (rq->polling) {
		rx_gather[gather_idx++] = skb;
		if (gather_idx == RX_BUNDLE_SIZE) {
			tdev->recv(tdev, rx_gather, RX_BUNDLE_SIZE);
			gather_idx = 0;
			rq->offload_bundles++;
		}
	} else
		offload_enqueue(rq, skb);

	return gather_idx;
}

/**
 *	restart_tx - check whether to restart suspended Tx queues
 *	@qs: the queue set to resume
 *
 *	Restarts suspended Tx queues of an SGE queue set if they have enough
 *	free resources to resume operation.
 */
static void restart_tx(struct sge_qset *qs)
{
	if (test_bit(TXQ_ETH, &qs->txq_stopped) &&
	    should_restart_tx(&qs->txq[TXQ_ETH]) &&
	    test_and_clear_bit(TXQ_ETH, &qs->txq_stopped)) {
		qs->txq[TXQ_ETH].restarts++;
		if (netif_running(qs->netdev))
			netif_tx_wake_queue(qs->tx_q);
	}

	if (test_bit(TXQ_OFLD, &qs->txq_stopped) &&
	    should_restart_tx(&qs->txq[TXQ_OFLD]) &&
	    test_and_clear_bit(TXQ_OFLD, &qs->txq_stopped)) {
		qs->txq[TXQ_OFLD].restarts++;
		tasklet_schedule(&qs->txq[TXQ_OFLD].qresume_tsk);
	}
	if (test_bit(TXQ_CTRL, &qs->txq_stopped) &&
	    should_restart_tx(&qs->txq[TXQ_CTRL]) &&
	    test_and_clear_bit(TXQ_CTRL, &qs->txq_stopped)) {
		qs->txq[TXQ_CTRL].restarts++;
		tasklet_schedule(&qs->txq[TXQ_CTRL].qresume_tsk);
	}
}

/**
 *	cxgb3_arp_process - process an ARP request probing a private IP address
 *	@adapter: the adapter
 *	@skb: the skbuff containing the ARP request
 *
 *	Check if the ARP request is probing the private IP address
 *	dedicated to iSCSI, generate an ARP reply if so.
 */
static void cxgb3_arp_process(struct port_info *pi, struct sk_buff *skb)
{
	struct net_device *dev = skb->dev;
	struct arphdr *arp;
	unsigned char *arp_ptr;
	unsigned char *sha;
	__be32 sip, tip;

	if (!dev)
		return;

	skb_reset_network_header(skb);
	arp = arp_hdr(skb);

	if (arp->ar_op != htons(ARPOP_REQUEST))
		return;

	arp_ptr = (unsigned char *)(arp + 1);
	sha = arp_ptr;
	arp_ptr += dev->addr_len;
	memcpy(&sip, arp_ptr, sizeof(sip));
	arp_ptr += sizeof(sip);
	arp_ptr += dev->addr_len;
	memcpy(&tip, arp_ptr, sizeof(tip));

	if (tip != pi->iscsi_ipv4addr)
		return;

	arp_send(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
		 pi->iscsic.mac_addr, sha);

}

static inline int is_arp(struct sk_buff *skb)
{
	return skb->protocol == htons(ETH_P_ARP);
}

static void cxgb3_process_iscsi_prov_pack(struct port_info *pi,
					struct sk_buff *skb)
{
	if (is_arp(skb)) {
		cxgb3_arp_process(pi, skb);
		return;
	}

	if (pi->iscsic.recv)
		pi->iscsic.recv(pi, skb);

}

/**
 *	rx_eth - process an ingress ethernet packet
 *	@adap: the adapter
 *	@rq: the response queue that received the packet
 *	@skb: the packet
 *	@pad: amount of padding at the start of the buffer
 *
 *	Process an ingress ethernet pakcet and deliver it to the stack.
 *	The padding is 2 if the packet was delivered in an Rx buffer and 0
 *	if it was immediate data in a response.
 */
static void rx_eth(struct adapter *adap, struct sge_rspq *rq,
		   struct sk_buff *skb, int pad, int lro)
{
	struct cpl_rx_pkt *p = (struct cpl_rx_pkt *)(skb->data + pad);
	struct sge_qset *qs = rspq_to_qset(rq);
	struct port_info *pi;

	skb_pull(skb, sizeof(*p) + pad);
	skb->protocol = eth_type_trans(skb, adap->port[p->iff]);
	pi = netdev_priv(skb->dev);
	if ((pi->rx_offload & T3_RX_CSUM) && p->csum_valid &&
	    p->csum == htons(0xffff) && !p->fragment) {
		qs->port_stats[SGE_PSTAT_RX_CSUM_GOOD]++;
		skb->ip_summed = CHECKSUM_UNNECESSARY;
	} else
		skb->ip_summed = CHECKSUM_NONE;
	skb_record_rx_queue(skb, qs - &adap->sge.qs[0]);

	if (unlikely(p->vlan_valid)) {
		struct vlan_group *grp = pi->vlan_grp;

		qs->port_stats[SGE_PSTAT_VLANEX]++;
		if (likely(grp))
			if (lro)
				vlan_gro_receive(&qs->napi, grp,
						 ntohs(p->vlan), skb);
			else {
				if (unlikely(pi->iscsic.flags)) {
					unsigned short vtag = ntohs(p->vlan) &
								VLAN_VID_MASK;
					skb->dev = vlan_group_get_device(grp,
									 vtag);
					cxgb3_process_iscsi_prov_pack(pi, skb);
				}
				__vlan_hwaccel_rx(skb, grp, ntohs(p->vlan),
					  	  rq->polling);
			}
		else
			dev_kfree_skb_any(skb);
	} else if (rq->polling) {
		if (lro)
			napi_gro_receive(&qs->napi, skb);
		else {
			if (unlikely(pi->iscsic.flags))
				cxgb3_process_iscsi_prov_pack(pi, skb);
			netif_receive_skb(skb);
		}
	} else
		netif_rx(skb);
}

static inline int is_eth_tcp(u32 rss)
{
	return G_HASHTYPE(ntohl(rss)) == RSS_HASH_4_TUPLE;
}

/**
 *	lro_add_page - add a page chunk to an LRO session
 *	@adap: the adapter
 *	@qs: the associated queue set
 *	@fl: the free list containing the page chunk to add
 *	@len: packet length
 *	@complete: Indicates the last fragment of a frame
 *
 *	Add a received packet contained in a page chunk to an existing LRO
 *	session.
 */
static void lro_add_page(struct adapter *adap, struct sge_qset *qs,
			 struct sge_fl *fl, int len, int complete)
{
	struct rx_sw_desc *sd = &fl->sdesc[fl->cidx];
	struct sk_buff *skb = NULL;
	struct cpl_rx_pkt *cpl;
	struct skb_frag_struct *rx_frag;
	int nr_frags;
	int offset = 0;

	if (!qs->nomem) {
		skb = napi_get_frags(&qs->napi);
		qs->nomem = !skb;
	}

	fl->credits--;

	pci_dma_sync_single_for_cpu(adap->pdev,
				    pci_unmap_addr(sd, dma_addr),
				    fl->buf_size - SGE_PG_RSVD,
				    PCI_DMA_FROMDEVICE);

	(*sd->pg_chunk.p_cnt)--;
	if (!*sd->pg_chunk.p_cnt)
		pci_unmap_page(adap->pdev,
			       sd->pg_chunk.mapping,
			       fl->alloc_size,
			       PCI_DMA_FROMDEVICE);

	if (!skb) {
		put_page(sd->pg_chunk.page);
		if (complete)
			qs->nomem = 0;
		return;
	}

	rx_frag = skb_shinfo(skb)->frags;
	nr_frags = skb_shinfo(skb)->nr_frags;

	if (!nr_frags) {
		offset = 2 + sizeof(struct cpl_rx_pkt);
		qs->lro_va = sd->pg_chunk.va + 2;
	}
	len -= offset;

	prefetch(qs->lro_va);

	rx_frag += nr_frags;
	rx_frag->page = sd->pg_chunk.page;
	rx_frag->page_offset = sd->pg_chunk.offset + offset;
	rx_frag->size = len;

	skb->len += len;
	skb->data_len += len;
	skb->truesize += len;
	skb_shinfo(skb)->nr_frags++;

	if (!complete)
		return;

	skb_record_rx_queue(skb, qs - &adap->sge.qs[0]);
	skb->ip_summed = CHECKSUM_UNNECESSARY;
	cpl = qs->lro_va;

	if (unlikely(cpl->vlan_valid)) {
		struct net_device *dev = qs->netdev;
		struct port_info *pi = netdev_priv(dev);
		struct vlan_group *grp = pi->vlan_grp;

		if (likely(grp != NULL)) {
			vlan_gro_frags(&qs->napi, grp, ntohs(cpl->vlan));
			return;
		}
	}
	napi_gro_frags(&qs->napi);
}

/**
 *	handle_rsp_cntrl_info - handles control information in a response
 *	@qs: the queue set corresponding to the response
 *	@flags: the response control flags
 *
 *	Handles the control information of an SGE response, such as GTS
 *	indications and completion credits for the queue set's Tx queues.
 *	HW coalesces credits, we don't do any extra SW coalescing.
 */
static inline void handle_rsp_cntrl_info(struct sge_qset *qs, u32 flags)
{
	unsigned int credits;

#if USE_GTS
	if (flags & F_RSPD_TXQ0_GTS)
		clear_bit(TXQ_RUNNING, &qs->txq[TXQ_ETH].flags);
#endif

	credits = G_RSPD_TXQ0_CR(flags);
	if (credits)
		qs->txq[TXQ_ETH].processed += credits;

	credits = G_RSPD_TXQ2_CR(flags);
	if (credits)
		qs->txq[TXQ_CTRL].processed += credits;

# if USE_GTS
	if (flags & F_RSPD_TXQ1_GTS)
		clear_bit(TXQ_RUNNING, &qs->txq[TXQ_OFLD].flags);
# endif
	credits = G_RSPD_TXQ1_CR(flags);
	if (credits)
		qs->txq[TXQ_OFLD].processed += credits;
}

/**
 *	check_ring_db - check if we need to ring any doorbells
 *	@adapter: the adapter
 *	@qs: the queue set whose Tx queues are to be examined
 *	@sleeping: indicates which Tx queue sent GTS
 *
 *	Checks if some of a queue set's Tx queues need to ring their doorbells
 *	to resume transmission after idling while they still have unprocessed
 *	descriptors.
 */
static void check_ring_db(struct adapter *adap, struct sge_qset *qs,
			  unsigned int sleeping)
{
	if (sleeping & F_RSPD_TXQ0_GTS) {
		struct sge_txq *txq = &qs->txq[TXQ_ETH];

		if (txq->cleaned + txq->in_use != txq->processed &&
		    !test_and_set_bit(TXQ_LAST_PKT_DB, &txq->flags)) {
			set_bit(TXQ_RUNNING, &txq->flags);
			t3_write_reg(adap, A_SG_KDOORBELL, F_SELEGRCNTX |
				     V_EGRCNTX(txq->cntxt_id));
		}
	}

	if (sleeping & F_RSPD_TXQ1_GTS) {
		struct sge_txq *txq = &qs->txq[TXQ_OFLD];

		if (txq->cleaned + txq->in_use != txq->processed &&
		    !test_and_set_bit(TXQ_LAST_PKT_DB, &txq->flags)) {
			set_bit(TXQ_RUNNING, &txq->flags);
			t3_write_reg(adap, A_SG_KDOORBELL, F_SELEGRCNTX |
				     V_EGRCNTX(txq->cntxt_id));
		}
	}
}

/**
 *	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 int is_new_response(const struct rsp_desc *r,
				  const struct sge_rspq *q)
{
	return (r->intr_gen & F_RSPD_GEN2) == q->gen;
}

static inline void clear_rspq_bufstate(struct sge_rspq * const q)
{
	q->pg_skb = NULL;
	q->rx_recycle_buf = 0;
}

#define RSPD_GTS_MASK  (F_RSPD_TXQ0_GTS | F_RSPD_TXQ1_GTS)
#define RSPD_CTRL_MASK (RSPD_GTS_MASK | \
			V_RSPD_TXQ0_CR(M_RSPD_TXQ0_CR) | \
			V_RSPD_TXQ1_CR(M_RSPD_TXQ1_CR) | \
			V_RSPD_TXQ2_CR(M_RSPD_TXQ2_CR))

/* How long to delay the next interrupt in case of memory shortage, in 0.1us. */
#define NOMEM_INTR_DELAY 2500

/**
 *	process_responses - process responses from an SGE response queue
 *	@adap: the adapter
 *	@qs: the queue set to which the response queue belongs
 *	@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 credits and other events
 *	for the queues that belong to the response queue's queue set.
 *	A negative budget is effectively unlimited.
 *
 *	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 adapter *adap, struct sge_qset *qs,
			     int budget)
{
	struct sge_rspq *q = &qs->rspq;
	struct rsp_desc *r = &q->desc[q->cidx];
	int budget_left = budget;
	unsigned int sleeping = 0;
	struct sk_buff *offload_skbs[RX_BUNDLE_SIZE];
	int ngathered = 0;

	q->next_holdoff = q->holdoff_tmr;

	while (likely(budget_left && is_new_response(r, q))) {
		int packet_complete, eth, ethpad = 2, lro = qs->lro_enabled;
		struct sk_buff *skb = NULL;
		u32 len, flags = ntohl(r->flags);
		__be32 rss_hi = *(const __be32 *)r,
		       rss_lo = r->rss_hdr.rss_hash_val;

		eth = r->rss_hdr.opcode == CPL_RX_PKT;

		if (unlikely(flags & F_RSPD_ASYNC_NOTIF)) {
			skb = alloc_skb(AN_PKT_SIZE, GFP_ATOMIC);
			if (!skb)
				goto no_mem;

			memcpy(__skb_put(skb, AN_PKT_SIZE), r, AN_PKT_SIZE);
			skb->data[0] = CPL_ASYNC_NOTIF;
			rss_hi = htonl(CPL_ASYNC_NOTIF << 24);
			q->async_notif++;
		} else if (flags & F_RSPD_IMM_DATA_VALID) {
			skb = get_imm_packet(r);
			if (unlikely(!skb)) {
no_mem:
				q->next_holdoff = NOMEM_INTR_DELAY;
				q->nomem++;
				/* consume one credit since we tried */
				budget_left--;
				break;
			}
			q->imm_data++;
			ethpad = 0;
		} else if ((len = ntohl(r->len_cq)) != 0) {
			struct sge_fl *fl;

			lro &= eth && is_eth_tcp(rss_hi);

			fl = (len & F_RSPD_FLQ) ? &qs->fl[1] : &qs->fl[0];
			if (fl->use_pages) {
				void *addr = fl->sdesc[fl->cidx].pg_chunk.va;

				prefetch(addr);
#if L1_CACHE_BYTES < 128
				prefetch(addr + L1_CACHE_BYTES);
#endif
				__refill_fl(adap, fl);
				if (lro > 0) {
					lro_add_page(adap, qs, fl,
						     G_RSPD_LEN(len),
						     flags & F_RSPD_EOP);
					 goto next_fl;
				}

				skb = get_packet_pg(adap, fl, q,
						    G_RSPD_LEN(len),
						    eth ?
						    SGE_RX_DROP_THRES : 0);
				q->pg_skb = skb;
			} else
				skb = get_packet(adap, fl, G_RSPD_LEN(len),
						 eth ? SGE_RX_DROP_THRES : 0);
			if (unlikely(!skb)) {
				if (!eth)
					goto no_mem;
				q->rx_drops++;
			} else if (unlikely(r->rss_hdr.opcode == CPL_TRACE_PKT))
				__skb_pull(skb, 2);
next_fl:
			if (++fl->cidx == fl->size)
				fl->cidx = 0;
		} else
			q->pure_rsps++;

		if (flags & RSPD_CTRL_MASK) {
			sleeping |= flags & RSPD_GTS_MASK;
			handle_rsp_cntrl_info(qs, flags);
		}

		r++;
		if (unlikely(++q->cidx == q->size)) {
			q->cidx = 0;
			q->gen ^= 1;
			r = q->desc;
		}
		prefetch(r);

		if (++q->credits >= (q->size / 4)) {
			refill_rspq(adap, q, q->credits);
			q->credits = 0;
		}

		packet_complete = flags &
				  (F_RSPD_EOP | F_RSPD_IMM_DATA_VALID |
				   F_RSPD_ASYNC_NOTIF);

		if (skb != NULL && packet_complete) {
			if (eth)
				rx_eth(adap, q, skb, ethpad, lro);
			else {
				q->offload_pkts++;
				/* Preserve the RSS info in csum & priority */
				skb->csum = rss_hi;
				skb->priority = rss_lo;
				ngathered = rx_offload(&adap->tdev, q, skb,
						       offload_skbs,
						       ngathered);
			}

			if (flags & F_RSPD_EOP)
				clear_rspq_bufstate(q);
		}
		--budget_left;
	}

	deliver_partial_bundle(&adap->tdev, q, offload_skbs, ngathered);

	if (sleeping)
		check_ring_db(adap, qs, sleeping);

	smp_mb();		/* commit Tx queue .processed updates */
	if (unlikely(qs->txq_stopped != 0))
		restart_tx(qs);

	budget -= budget_left;
	return budget;
}

static inline int is_pure_response(const struct rsp_desc *r)
{
	__be32 n = r->flags & htonl(F_RSPD_ASYNC_NOTIF | F_RSPD_IMM_DATA_VALID);

	return (n | r->len_cq) == 0;
}

/**
 *	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.
 */
static int napi_rx_handler(struct napi_struct *napi, int budget)
{
	struct sge_qset *qs = container_of(napi, struct sge_qset, napi);
	struct adapter *adap = qs->adap;
	int work_done = process_responses(adap, qs, budget);

	if (likely(work_done < budget)) {
		napi_complete(napi);

		/*
		 * Because we don't atomically flush the following
		 * write it is possible that in very rare cases it can
		 * reach the device in a way that races with a new
		 * response being written plus an error interrupt
		 * causing the NAPI interrupt handler below to return
		 * unhandled status to the OS.  To protect against
		 * this would require flushing the write and doing
		 * both the write and the flush with interrupts off.
		 * Way too expensive and unjustifiable given the
		 * rarity of the race.
		 *
		 * The race cannot happen at all with MSI-X.
		 */
		t3_write_reg(adap, A_SG_GTS, V_RSPQ(qs->rspq.cntxt_id) |
			     V_NEWTIMER(qs->rspq.next_holdoff) |
			     V_NEWINDEX(qs->rspq.cidx));
	}
	return work_done;
}

/*
 * Returns true if the device is already scheduled for polling.
 */
static inline int napi_is_scheduled(struct napi_struct *napi)
{
	return test_bit(NAPI_STATE_SCHED, &napi->state);
}

/**
 *	process_pure_responses - process pure responses from a response queue
 *	@adap: the adapter
 *	@qs: the queue set owning the response queue
 *	@r: the first pure response to process
 *
 *	A simpler version of process_responses() that handles only pure (i.e.,
 *	non data-carrying) responses.  Such respones are too light-weight to
 *	justify calling a softirq under NAPI, so we handle them specially in
 *	the interrupt handler.  The function is called with a pointer to a
 *	response, which the caller must ensure is a valid pure response.
 *
 *	Returns 1 if it encounters a valid data-carrying response, 0 otherwise.
 */
static int process_pure_responses(struct adapter *adap, struct sge_qset *qs,
				  struct rsp_desc *r)
{
	struct sge_rspq *q = &qs->rspq;
	unsigned int sleeping = 0;

	do {
		u32 flags = ntohl(r->flags);

		r++;
		if (unlikely(++q->cidx == q->size)) {
			q->cidx = 0;
			q->gen ^= 1;
			r = q->desc;
		}
		prefetch(r);

		if (flags & RSPD_CTRL_MASK) {
			sleeping |= flags & RSPD_GTS_MASK;
			handle_rsp_cntrl_info(qs, flags);
		}

		q->pure_rsps++;
		if (++q->credits >= (q->size / 4)) {
			refill_rspq(adap, q, q->credits);
			q->credits = 0;
		}
	} while (is_new_response(r, q) && is_pure_response(r));

	if (sleeping)
		check_ring_db(adap, qs, sleeping);

	smp_mb();		/* commit Tx queue .processed updates */
	if (unlikely(qs->txq_stopped != 0))
		restart_tx(qs);

	return is_new_response(r, q);
}

/**
 *	handle_responses - decide what to do with new responses in NAPI mode
 *	@adap: the adapter
 *	@q: the response queue
 *
 *	This is used by the NAPI interrupt handlers to decide what to do with
 *	new SGE responses.  If there are no new responses it returns -1.  If
 *	there are new responses and they are pure (i.e., non-data carrying)
 *	it handles them straight in hard interrupt context as they are very
 *	cheap and don't deliver any packets.  Finally, if there are any data
 *	signaling responses it schedules the NAPI handler.  Returns 1 if it
 *	schedules NAPI, 0 if all new responses were pure.
 *
 *	The caller must ascertain NAPI is not already running.
 */
static inline int handle_responses(struct adapter *adap, struct sge_rspq *q)
{
	struct sge_qset *qs = rspq_to_qset(q);
	struct rsp_desc *r = &q->desc[q->cidx];

	if (!is_new_response(r, q))
		return -1;
	if (is_pure_response(r) && process_pure_responses(adap, qs, r) == 0) {
		t3_write_reg(adap, A_SG_GTS, V_RSPQ(q->cntxt_id) |
			     V_NEWTIMER(q->holdoff_tmr) | V_NEWINDEX(q->cidx));
		return 0;
	}
	napi_schedule(&qs->napi);
	return 1;
}

/*
 * The MSI-X interrupt handler for an SGE response queue for the non-NAPI case
 * (i.e., response queue serviced in hard interrupt).
 */
irqreturn_t t3_sge_intr_msix(int irq, void *cookie)
{
	struct sge_qset *qs = cookie;
	struct adapter *adap = qs->adap;
	struct sge_rspq *q = &qs->rspq;

	spin_lock(&q->lock);
	if (process_responses(adap, qs, -1) == 0)
		q->unhandled_irqs++;
	t3_write_reg(adap, A_SG_GTS, V_RSPQ(q->cntxt_id) |
		     V_NEWTIMER(q->next_holdoff) | V_NEWINDEX(q->cidx));
	spin_unlock(&q->lock);
	return IRQ_HANDLED;
}

/*
 * The MSI-X interrupt handler for an SGE response queue for the NAPI case
 * (i.e., response queue serviced by NAPI polling).
 */
static irqreturn_t t3_sge_intr_msix_napi(int irq, void *cookie)
{
	struct sge_qset *qs = cookie;
	struct sge_rspq *q = &qs->rspq;

	spin_lock(&q->lock);

	if (handle_responses(qs->adap, q) < 0)
		q->unhandled_irqs++;
	spin_unlock(&q->lock);
	return IRQ_HANDLED;
}

/*
 * The non-NAPI MSI interrupt handler.  This needs to handle data events from
 * SGE response queues as well as error and other async events as they all use
 * the same MSI vector.  We use one SGE response queue per port in this mode
 * and protect all response queues with queue 0's lock.
 */
static irqreturn_t t3_intr_msi(int irq, void *cookie)
{
	int new_packets = 0;
	struct adapter *adap = cookie;
	struct sge_rspq *q = &adap->sge.qs[0].rspq;

	spin_lock(&q->lock);

	if (process_responses(adap, &adap->sge.qs[0], -1)) {
		t3_write_reg(adap, A_SG_GTS, V_RSPQ(q->cntxt_id) |
			     V_NEWTIMER(q->next_holdoff) | V_NEWINDEX(q->cidx));
		new_packets = 1;
	}

	if (adap->params.nports == 2 &&
	    process_responses(adap, &adap->sge.qs[1], -1)) {
		struct sge_rspq *q1 = &adap->sge.qs[1].rspq;

		t3_write_reg(adap, A_SG_GTS, V_RSPQ(q1->cntxt_id) |
			     V_NEWTIMER(q1->next_holdoff) |
			     V_NEWINDEX(q1->cidx));
		new_packets = 1;
	}

	if (!new_packets && t3_slow_intr_handler(adap) == 0)
		q->unhandled_irqs++;

	spin_unlock(&q->lock);
	return IRQ_HANDLED;
}

static int rspq_check_napi(struct sge_qset *qs)
{
	struct sge_rspq *q = &qs->rspq;

	if (!napi_is_scheduled(&qs->napi) &&
	    is_new_response(&q->desc[q->cidx], q)) {
		napi_schedule(&qs->napi);
		return 1;
	}
	return 0;
}

/*
 * The MSI interrupt handler for the NAPI case (i.e., response queues serviced
 * by NAPI polling).  Handles data events from SGE response queues as well as
 * error and other async events as they all use the same MSI vector.  We use
 * one SGE response queue per port in this mode and protect all response
 * queues with queue 0's lock.
 */
static irqreturn_t t3_intr_msi_napi(int irq, void *cookie)
{
	int new_packets;
	struct adapter *adap = cookie;
	struct sge_rspq *q = &adap->sge.qs[0].rspq;

	spin_lock(&q->lock);

	new_packets = rspq_check_napi(&adap->sge.qs[0]);
	if (adap->params.nports == 2)
		new_packets += rspq_check_napi(&adap->sge.qs[1]);
	if (!new_packets && t3_slow_intr_handler(adap) == 0)
		q->unhandled_irqs++;

	spin_unlock(&q->lock);
	return IRQ_HANDLED;
}

/*
 * A helper function that processes responses and issues GTS.
 */
static inline int process_responses_gts(struct adapter *adap,
					struct sge_rspq *rq)
{
	int work;

	work = process_responses(adap, rspq_to_qset(rq), -1);
	t3_write_reg(adap, A_SG_GTS, V_RSPQ(rq->cntxt_id) |
		     V_NEWTIMER(rq->next_holdoff) | V_NEWINDEX(rq->cidx));
	return work;
}

/*
 * The legacy INTx interrupt handler.  This needs to handle data events from
 * SGE response queues as well as error and other async events as they all use
 * the same interrupt pin.  We use one SGE response queue per port in this mode
 * and protect all response queues with queue 0's lock.
 */
static irqreturn_t t3_intr(int irq, void *cookie)
{
	int work_done, w0, w1;
	struct adapter *adap = cookie;
	struct sge_rspq *q0 = &adap->sge.qs[0].rspq;
	struct sge_rspq *q1 = &adap->sge.qs[1].rspq;

	spin_lock(&q0->lock);

	w0 = is_new_response(&q0->desc[q0->cidx], q0);
	w1 = adap->params.nports == 2 &&
	    is_new_response(&q1->desc[q1->cidx], q1);

	if (likely(w0 | w1)) {
		t3_write_reg(adap, A_PL_CLI, 0);
		t3_read_reg(adap, A_PL_CLI);	/* flush */

		if (likely(w0))
			process_responses_gts(adap, q0);

		if (w1)
			process_responses_gts(adap, q1);

		work_done = w0 | w1;
	} else
		work_done = t3_slow_intr_handler(adap);

	spin_unlock(&q0->lock);
	return IRQ_RETVAL(work_done != 0);
}

/*
 * Interrupt handler for legacy INTx interrupts for T3B-based cards.
 * Handles data events from SGE response queues as well as error and other
 * async events as they all use the same interrupt pin.  We use one SGE
 * response queue per port in this mode and protect all response queues with
 * queue 0's lock.
 */
static irqreturn_t t3b_intr(int irq, void *cookie)
{
	u32 map;
	struct adapter *adap = cookie;
	struct sge_rspq *q0 = &adap->sge.qs[0].rspq;

	t3_write_reg(adap, A_PL_CLI, 0);
	map = t3_read_reg(adap, A_SG_DATA_INTR);

	if (unlikely(!map))	/* shared interrupt, most likely */
		return IRQ_NONE;

	spin_lock(&q0->lock);

	if (unlikely(map & F_ERRINTR))
		t3_slow_intr_handler(adap);

	if (likely(map & 1))
		process_responses_gts(adap, q0);

	if (map & 2)
		process_responses_gts(adap, &adap->sge.qs[1].rspq);

	spin_unlock(&q0->lock);
	return IRQ_HANDLED;
}

/*
 * NAPI interrupt handler for legacy INTx interrupts for T3B-based cards.
 * Handles data events from SGE response queues as well as error and other
 * async events as they all use the same interrupt pin.  We use one SGE
 * response queue per port in this mode and protect all response queues with
 * queue 0's lock.
 */
static irqreturn_t t3b_intr_napi(int irq, void *cookie)
{
	u32 map;
	struct adapter *adap = cookie;
	struct sge_qset *qs0 = &adap->sge.qs[0];
	struct sge_rspq *q0 = &qs0->rspq;

	t3_write_reg(adap, A_PL_CLI, 0);
	map = t3_read_reg(adap, A_SG_DATA_INTR);

	if (unlikely(!map))	/* shared interrupt, most likely */
		return IRQ_NONE;

	spin_lock(&q0->lock);

	if (unlikely(map & F_ERRINTR))
		t3_slow_intr_handler(adap);

	if (likely(map & 1))
		napi_schedule(&qs0->napi);

	if (map & 2)
		napi_schedule(&adap->sge.qs[1].napi);

	spin_unlock(&q0->lock);
	return IRQ_HANDLED;
}

/**
 *	t3_intr_handler - select the top-level interrupt handler
 *	@adap: the adapter
 *	@polling: whether using NAPI to service response queues
 *
 *	Selects the top-level interrupt handler based on the type of interrupts
 *	(MSI-X, MSI, or legacy) and whether NAPI will be used to service the
 *	response queues.
 */
irq_handler_t t3_intr_handler(struct adapter *adap, int polling)
{
	if (adap->flags & USING_MSIX)
		return polling ? t3_sge_intr_msix_napi : t3_sge_intr_msix;
	if (adap->flags & USING_MSI)
		return polling ? t3_intr_msi_napi : t3_intr_msi;
	if (adap->params.rev > 0)
		return polling ? t3b_intr_napi : t3b_intr;
	return t3_intr;
}

#define SGE_PARERR (F_CPPARITYERROR | F_OCPARITYERROR | F_RCPARITYERROR | \
		    F_IRPARITYERROR | V_ITPARITYERROR(M_ITPARITYERROR) | \
		    V_FLPARITYERROR(M_FLPARITYERROR) | F_LODRBPARITYERROR | \
		    F_HIDRBPARITYERROR | F_LORCQPARITYERROR | \
		    F_HIRCQPARITYERROR)
#define SGE_FRAMINGERR (F_UC_REQ_FRAMINGERROR | F_R_REQ_FRAMINGERROR)
#define SGE_FATALERR (SGE_PARERR | SGE_FRAMINGERR | F_RSPQCREDITOVERFOW | \
		      F_RSPQDISABLED)

/**
 *	t3_sge_err_intr_handler - SGE async event interrupt handler
 *	@adapter: the adapter
 *
 *	Interrupt handler for SGE asynchronous (non-data) events.
 */
void t3_sge_err_intr_handler(struct adapter *adapter)
{
	unsigned int v, status = t3_read_reg(adapter, A_SG_INT_CAUSE) &
				 ~F_FLEMPTY;

	if (status & SGE_PARERR)
		CH_ALERT(adapter, "SGE parity error (0x%x)\n",
			 status & SGE_PARERR);
	if (status & SGE_FRAMINGERR)
		CH_ALERT(adapter, "SGE framing error (0x%x)\n",
			 status & SGE_FRAMINGERR);

	if (status & F_RSPQCREDITOVERFOW)
		CH_ALERT(adapter, "SGE response queue credit overflow\n");

	if (status & F_RSPQDISABLED) {
		v = t3_read_reg(adapter, A_SG_RSPQ_FL_STATUS);

		CH_ALERT(adapter,
			 "packet delivered to disabled response queue "
			 "(0x%x)\n", (v >> S_RSPQ0DISABLED) & 0xff);
	}

	if (status & (F_HIPIODRBDROPERR | F_LOPIODRBDROPERR))
		CH_ALERT(adapter, "SGE dropped %s priority doorbell\n",
			 status & F_HIPIODRBDROPERR ? "high" : "lo");

	t3_write_reg(adapter, A_SG_INT_CAUSE, status);
	if (status &  SGE_FATALERR)
		t3_fatal_err(adapter);
}

/**
 *	sge_timer_tx - perform periodic maintenance of an SGE qset
 *	@data: the SGE queue set to maintain
 *
 *	Runs periodically from a timer to perform maintenance of an SGE queue
 *	set.  It performs two tasks:
 *
 *	Cleans up any completed Tx descriptors that may still be pending.
 *	Normal descriptor cleanup happens when new packets are added to a Tx
 *	queue so this timer is relatively infrequent and does any cleanup only
 *	if the Tx queue has not seen any new packets in a while.  We make a
 *	best effort attempt to reclaim descriptors, in that we don't wait
 *	around if we cannot get a queue's lock (which most likely is because
 *	someone else is queueing new packets and so will also handle the clean
 *	up).  Since control queues use immediate data exclusively we don't
 *	bother cleaning them up here.
 *
 */
static void sge_timer_tx(unsigned long data)
{
	struct sge_qset *qs = (struct sge_qset *)data;
	struct port_info *pi = netdev_priv(qs->netdev);
	struct adapter *adap = pi->adapter;
	unsigned int tbd[SGE_TXQ_PER_SET] = {0, 0};
	unsigned long next_period;

	if (__netif_tx_trylock(qs->tx_q)) {
                tbd[TXQ_ETH] = reclaim_completed_tx(adap, &qs->txq[TXQ_ETH],
                                                     TX_RECLAIM_TIMER_CHUNK);
		__netif_tx_unlock(qs->tx_q);
	}

	if (spin_trylock(&qs->txq[TXQ_OFLD].lock)) {
		tbd[TXQ_OFLD] = reclaim_completed_tx(adap, &qs->txq[TXQ_OFLD],
						     TX_RECLAIM_TIMER_CHUNK);
		spin_unlock(&qs->txq[TXQ_OFLD].lock);
	}

	next_period = TX_RECLAIM_PERIOD >>
                      (max(tbd[TXQ_ETH], tbd[TXQ_OFLD]) /
                      TX_RECLAIM_TIMER_CHUNK);
	mod_timer(&qs->tx_reclaim_timer, jiffies + next_period);
}

/*
 *	sge_timer_rx - perform periodic maintenance of an SGE qset
 *	@data: the SGE queue set to maintain
 *
 *	a) Replenishes Rx queues that have run out due to memory shortage.
 *	Normally new Rx buffers are added when existing ones are consumed but
 *	when out of memory a queue can become empty.  We try to add only a few
 *	buffers here, the queue will be replenished fully as these new buffers
 *	are used up if memory shortage has subsided.
 *
 *	b) Return coalesced response queue credits in case a response queue is
 *	starved.
 *
 */
static void sge_timer_rx(unsigned long data)
{
	spinlock_t *lock;
	struct sge_qset *qs = (struct sge_qset *)data;
	struct port_info *pi = netdev_priv(qs->netdev);
	struct adapter *adap = pi->adapter;
	u32 status;

	lock = adap->params.rev > 0 ?
	       &qs->rspq.lock : &adap->sge.qs[0].rspq.lock;

	if (!spin_trylock_irq(lock))
		goto out;

	if (napi_is_scheduled(&qs->napi))
		goto unlock;

	if (adap->params.rev < 4) {
		status = t3_read_reg(adap, A_SG_RSPQ_FL_STATUS);

		if (status & (1 << qs->rspq.cntxt_id)) {
			qs->rspq.starved++;
			if (qs->rspq.credits) {
				qs->rspq.credits--;
				refill_rspq(adap, &qs->rspq, 1);
				qs->rspq.restarted++;
				t3_write_reg(adap, A_SG_RSPQ_FL_STATUS,
					     1 << qs->rspq.cntxt_id);
			}
		}
	}

	if (qs->fl[0].credits < qs->fl[0].size)
		__refill_fl(adap, &qs->fl[0]);
	if (qs->fl[1].credits < qs->fl[1].size)
		__refill_fl(adap, &qs->fl[1]);

unlock:
	spin_unlock_irq(lock);
out:
	mod_timer(&qs->rx_reclaim_timer, jiffies + RX_RECLAIM_PERIOD);
}

/**
 *	t3_update_qset_coalesce - update coalescing settings for a queue set
 *	@qs: the SGE queue set
 *	@p: new queue set parameters
 *
 *	Update the coalescing settings for an SGE queue set.  Nothing is done
 *	if the queue set is not initialized yet.
 */
void t3_update_qset_coalesce(struct sge_qset *qs, const struct qset_params *p)
{
	qs->rspq.holdoff_tmr = max(p->coalesce_usecs * 10, 1U);/* can't be 0 */
	qs->rspq.polling = p->polling;
	qs->napi.poll = p->polling ? napi_rx_handler : ofld_poll;
}

/**
 *	t3_sge_alloc_qset - initialize an SGE queue set
 *	@adapter: the adapter
 *	@id: the queue set id
 *	@nports: how many Ethernet ports will be using this queue set
 *	@irq_vec_idx: the IRQ vector index for response queue interrupts
 *	@p: configuration parameters for this queue set
 *	@ntxq: number of Tx queues for the queue set
 *	@netdev: net device associated with this queue set
 *	@netdevq: net device TX queue associated with this queue set
 *
 *	Allocate resources and initialize an SGE queue set.  A queue set
 *	comprises a response queue, two Rx free-buffer queues, and up to 3
 *	Tx queues.  The Tx queues are assigned roles in the order Ethernet
 *	queue, offload queue, and control queue.
 */
int t3_sge_alloc_qset(struct adapter *adapter, unsigned int id, int nports,
		      int irq_vec_idx, const struct qset_params *p,
		      int ntxq, struct net_device *dev,
		      struct netdev_queue *netdevq)
{
	int i, avail, ret = -ENOMEM;
	struct sge_qset *q = &adapter->sge.qs[id];

	init_qset_cntxt(q, id);
	setup_timer(&q->tx_reclaim_timer, sge_timer_tx, (unsigned long)q);
	setup_timer(&q->rx_reclaim_timer, sge_timer_rx, (unsigned long)q);

	q->fl[0].desc = alloc_ring(adapter->pdev, p->fl_size,
				   sizeof(struct rx_desc),
				   sizeof(struct rx_sw_desc),
				   &q->fl[0].phys_addr, &q->fl[0].sdesc);
	if (!q->fl[0].desc)
		goto err;

	q->fl[1].desc = alloc_ring(adapter->pdev, p->jumbo_size,
				   sizeof(struct rx_desc),
				   sizeof(struct rx_sw_desc),
				   &q->fl[1].phys_addr, &q->fl[1].sdesc);
	if (!q->fl[1].desc)
		goto err;

	q->rspq.desc = alloc_ring(adapter->pdev, p->rspq_size,
				  sizeof(struct rsp_desc), 0,
				  &q->rspq.phys_addr, NULL);
	if (!q->rspq.desc)
		goto err;

	for (i = 0; i < ntxq; ++i) {
		/*
		 * The control queue always uses immediate data so does not
		 * need to keep track of any sk_buffs.
		 */
		size_t sz = i == TXQ_CTRL ? 0 : sizeof(struct tx_sw_desc);

		q->txq[i].desc = alloc_ring(adapter->pdev, p->txq_size[i],
					    sizeof(struct tx_desc), sz,
					    &q->txq[i].phys_addr,
					    &q->txq[i].sdesc);
		if (!q->txq[i].desc)
			goto err;

		q->txq[i].gen = 1;
		q->txq[i].size = p->txq_size[i];
		spin_lock_init(&q->txq[i].lock);
		skb_queue_head_init(&q->txq[i].sendq);
	}

	tasklet_init(&q->txq[TXQ_OFLD].qresume_tsk, restart_offloadq,
		     (unsigned long)q);
	tasklet_init(&q->txq[TXQ_CTRL].qresume_tsk, restart_ctrlq,
		     (unsigned long)q);

	q->fl[0].gen = q->fl[1].gen = 1;
	q->fl[0].size = p->fl_size;
	q->fl[1].size = p->jumbo_size;

	q->rspq.gen = 1;
	q->rspq.size = p->rspq_size;
	spin_lock_init(&q->rspq.lock);
	skb_queue_head_init(&q->rspq.rx_queue);

	q->txq[TXQ_ETH].stop_thres = nports *
	    flits_to_desc(sgl_len(MAX_SKB_FRAGS + 1) + 3);

#if FL0_PG_CHUNK_SIZE > 0
	q->fl[0].buf_size = FL0_PG_CHUNK_SIZE;
#else
	q->fl[0].buf_size = SGE_RX_SM_BUF_SIZE + sizeof(struct cpl_rx_data);
#endif
#if FL1_PG_CHUNK_SIZE > 0
	q->fl[1].buf_size = FL1_PG_CHUNK_SIZE;
#else
	q->fl[1].buf_size = is_offload(adapter) ?
		(16 * 1024) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
		MAX_FRAME_SIZE + 2 + sizeof(struct cpl_rx_pkt);
#endif

	q->fl[0].use_pages = FL0_PG_CHUNK_SIZE > 0;
	q->fl[1].use_pages = FL1_PG_CHUNK_SIZE > 0;
	q->fl[0].order = FL0_PG_ORDER;
	q->fl[1].order = FL1_PG_ORDER;
	q->fl[0].alloc_size = FL0_PG_ALLOC_SIZE;
	q->fl[1].alloc_size = FL1_PG_ALLOC_SIZE;

	spin_lock_irq(&adapter->sge.reg_lock);

	/* FL threshold comparison uses < */
	ret = t3_sge_init_rspcntxt(adapter, q->rspq.cntxt_id, irq_vec_idx,
				   q->rspq.phys_addr, q->rspq.size,
				   q->fl[0].buf_size - SGE_PG_RSVD, 1, 0);
	if (ret)
		goto err_unlock;

	for (i = 0; i < SGE_RXQ_PER_SET; ++i) {
		ret = t3_sge_init_flcntxt(adapter, q->fl[i].cntxt_id, 0,
					  q->fl[i].phys_addr, q->fl[i].size,
					  q->fl[i].buf_size - SGE_PG_RSVD,
					  p->cong_thres, 1, 0);
		if (ret)
			goto err_unlock;
	}

	ret = t3_sge_init_ecntxt(adapter, q->txq[TXQ_ETH].cntxt_id, USE_GTS,
				 SGE_CNTXT_ETH, id, q->txq[TXQ_ETH].phys_addr,
				 q->txq[TXQ_ETH].size, q->txq[TXQ_ETH].token,
				 1, 0);
	if (ret)
		goto err_unlock;

	if (ntxq > 1) {
		ret = t3_sge_init_ecntxt(adapter, q->txq[TXQ_OFLD].cntxt_id,
					 USE_GTS, SGE_CNTXT_OFLD, id,
					 q->txq[TXQ_OFLD].phys_addr,
					 q->txq[TXQ_OFLD].size, 0, 1, 0);
		if (ret)
			goto err_unlock;
	}

	if (ntxq > 2) {
		ret = t3_sge_init_ecntxt(adapter, q->txq[TXQ_CTRL].cntxt_id, 0,
					 SGE_CNTXT_CTRL, id,
					 q->txq[TXQ_CTRL].phys_addr,
					 q->txq[TXQ_CTRL].size,
					 q->txq[TXQ_CTRL].token, 1, 0);
		if (ret)
			goto err_unlock;
	}

	spin_unlock_irq(&adapter->sge.reg_lock);

	q->adap = adapter;
	q->netdev = dev;
	q->tx_q = netdevq;
	t3_update_qset_coalesce(q, p);

	avail = refill_fl(adapter, &q->fl[0], q->fl[0].size,
			  GFP_KERNEL | __GFP_COMP);
	if (!avail) {
		CH_ALERT(adapter, "free list queue 0 initialization failed\n");
		goto err;
	}
	if (avail < q->fl[0].size)
		CH_WARN(adapter, "free list queue 0 enabled with %d credits\n",
			avail);

	avail = refill_fl(adapter, &q->fl[1], q->fl[1].size,
			  GFP_KERNEL | __GFP_COMP);
	if (avail < q->fl[1].size)
		CH_WARN(adapter, "free list queue 1 enabled with %d credits\n",
			avail);
	refill_rspq(adapter, &q->rspq, q->rspq.size - 1);

	t3_write_reg(adapter, A_SG_GTS, V_RSPQ(q->rspq.cntxt_id) |
		     V_NEWTIMER(q->rspq.holdoff_tmr));

	return 0;

err_unlock:
	spin_unlock_irq(&adapter->sge.reg_lock);
err:
	t3_free_qset(adapter, q);
	return ret;
}

/**
 *      t3_start_sge_timers - start SGE timer call backs
 *      @adap: the adapter
 *
 *      Starts each SGE queue set's timer call back
 */
void t3_start_sge_timers(struct adapter *adap)
{
	int i;

	for (i = 0; i < SGE_QSETS; ++i) {
		struct sge_qset *q = &adap->sge.qs[i];

	if (q->tx_reclaim_timer.function)
		mod_timer(&q->tx_reclaim_timer, jiffies + TX_RECLAIM_PERIOD);

	if (q->rx_reclaim_timer.function)
		mod_timer(&q->rx_reclaim_timer, jiffies + RX_RECLAIM_PERIOD);
	}
}

/**
 *	t3_stop_sge_timers - stop SGE timer call backs
 *	@adap: the adapter
 *
 *	Stops each SGE queue set's timer call back
 */
void t3_stop_sge_timers(struct adapter *adap)
{
	int i;

	for (i = 0; i < SGE_QSETS; ++i) {
		struct sge_qset *q = &adap->sge.qs[i];

		if (q->tx_reclaim_timer.function)
			del_timer_sync(&q->tx_reclaim_timer);
		if (q->rx_reclaim_timer.function)
			del_timer_sync(&q->rx_reclaim_timer);
	}
}

/**
 *	t3_free_sge_resources - free SGE resources
 *	@adap: the adapter
 *
 *	Frees resources used by the SGE queue sets.
 */
void t3_free_sge_resources(struct adapter *adap)
{
	int i;

	for (i = 0; i < SGE_QSETS; ++i)
		t3_free_qset(adap, &adap->sge.qs[i]);
}

/**
 *	t3_sge_start - enable SGE
 *	@adap: the adapter
 *
 *	Enables the SGE for DMAs.  This is the last step in starting packet
 *	transfers.
 */
void t3_sge_start(struct adapter *adap)
{
	t3_set_reg_field(adap, A_SG_CONTROL, F_GLOBALENABLE, F_GLOBALENABLE);
}

/**
 *	t3_sge_stop - disable SGE operation
 *	@adap: the adapter
 *
 *	Disables the DMA engine.  This can be called in emeregencies (e.g.,
 *	from error interrupts) or from normal process context.  In the latter
 *	case it also disables any pending queue restart tasklets.  Note that
 *	if it is called in interrupt context it cannot disable the restart
 *	tasklets as it cannot wait, however the tasklets will have no effect
 *	since the doorbells are disabled and the driver will call this again
 *	later from process context, at which time the tasklets will be stopped
 *	if they are still running.
 */
void t3_sge_stop(struct adapter *adap)
{
	t3_set_reg_field(adap, A_SG_CONTROL, F_GLOBALENABLE, 0);
	if (!in_interrupt()) {
		int i;

		for (i = 0; i < SGE_QSETS; ++i) {
			struct sge_qset *qs = &adap->sge.qs[i];

			tasklet_kill(&qs->txq[TXQ_OFLD].qresume_tsk);
			tasklet_kill(&qs->txq[TXQ_CTRL].qresume_tsk);
		}
	}
}

/**
 *	t3_sge_init - initialize SGE
 *	@adap: the adapter
 *	@p: the SGE parameters
 *
 *	Performs SGE initialization needed every time after a chip reset.
 *	We do not initialize any of the queue sets here, instead the driver
 *	top-level must request those individually.  We also do not enable DMA
 *	here, that should be done after the queues have been set up.
 */
void t3_sge_init(struct adapter *adap, struct sge_params *p)
{
	unsigned int ctrl, ups = ffs(pci_resource_len(adap->pdev, 2) >> 12);

	ctrl = F_DROPPKT | V_PKTSHIFT(2) | F_FLMODE | F_AVOIDCQOVFL |
	    F_CQCRDTCTRL | F_CONGMODE | F_TNLFLMODE | F_FATLPERREN |
	    V_HOSTPAGESIZE(PAGE_SHIFT - 11) | F_BIGENDIANINGRESS |
	    V_USERSPACESIZE(ups ? ups - 1 : 0) | F_ISCSICOALESCING;
#if SGE_NUM_GENBITS == 1
	ctrl |= F_EGRGENCTRL;
#endif
	if (adap->params.rev > 0) {
		if (!(adap->flags & (USING_MSIX | USING_MSI)))
			ctrl |= F_ONEINTMULTQ | F_OPTONEINTMULTQ;
	}
	t3_write_reg(adap, A_SG_CONTROL, ctrl);
	t3_write_reg(adap, A_SG_EGR_RCQ_DRB_THRSH, V_HIRCQDRBTHRSH(512) |
		     V_LORCQDRBTHRSH(512));
	t3_write_reg(adap, A_SG_TIMER_TICK, core_ticks_per_usec(adap) / 10);
	t3_write_reg(adap, A_SG_CMDQ_CREDIT_TH, V_THRESHOLD(32) |
		     V_TIMEOUT(200 * core_ticks_per_usec(adap)));
	t3_write_reg(adap, A_SG_HI_DRB_HI_THRSH,
		     adap->params.rev < T3_REV_C ? 1000 : 500);
	t3_write_reg(adap, A_SG_HI_DRB_LO_THRSH, 256);
	t3_write_reg(adap, A_SG_LO_DRB_HI_THRSH, 1000);
	t3_write_reg(adap, A_SG_LO_DRB_LO_THRSH, 256);
	t3_write_reg(adap, A_SG_OCO_BASE, V_BASE1(0xfff));
	t3_write_reg(adap, A_SG_DRB_PRI_THRESH, 63 * 1024);
}

/**
 *	t3_sge_prep - one-time SGE initialization
 *	@adap: the associated adapter
 *	@p: SGE parameters
 *
 *	Performs one-time initialization of SGE SW state.  Includes determining
 *	defaults for the assorted SGE parameters, which admins can change until
 *	they are used to initialize the SGE.
 */
void t3_sge_prep(struct adapter *adap, struct sge_params *p)
{
	int i;

	p->max_pkt_size = (16 * 1024) - sizeof(struct cpl_rx_data) -
	    SKB_DATA_ALIGN(sizeof(struct skb_shared_info));

	for (i = 0; i < SGE_QSETS; ++i) {
		struct qset_params *q = p->qset + i;

		q->polling = adap->params.rev > 0;
		q->coalesce_usecs = 5;
		q->rspq_size = 1024;
		q->fl_size = 1024;
 		q->jumbo_size = 512;
		q->txq_size[TXQ_ETH] = 1024;
		q->txq_size[TXQ_OFLD] = 1024;
		q->txq_size[TXQ_CTRL] = 256;
		q->cong_thres = 0;
	}

	spin_lock_init(&adap->sge.reg_lock);
}

/**
 *	t3_get_desc - dump an SGE descriptor for debugging purposes
 *	@qs: the queue set
 *	@qnum: identifies the specific queue (0..2: Tx, 3:response, 4..5: Rx)
 *	@idx: the descriptor index in the queue
 *	@data: where to dump the descriptor contents
 *
 *	Dumps the contents of a HW descriptor of an SGE queue.  Returns the
 *	size of the descriptor.
 */
int t3_get_desc(const struct sge_qset *qs, unsigned int qnum, unsigned int idx,
		unsigned char *data)
{
	if (qnum >= 6)
		return -EINVAL;

	if (qnum < 3) {
		if (!qs->txq[qnum].desc || idx >= qs->txq[qnum].size)
			return -EINVAL;
		memcpy(data, &qs->txq[qnum].desc[idx], sizeof(struct tx_desc));
		return sizeof(struct tx_desc);
	}

	if (qnum == 3) {
		if (!qs->rspq.desc || idx >= qs->rspq.size)
			return -EINVAL;
		memcpy(data, &qs->rspq.desc[idx], sizeof(struct rsp_desc));
		return sizeof(struct rsp_desc);
	}

	qnum -= 4;
	if (!qs->fl[qnum].desc || idx >= qs->fl[qnum].size)
		return -EINVAL;
	memcpy(data, &qs->fl[qnum].desc[idx], sizeof(struct rx_desc));
	return sizeof(struct rx_desc);
}