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
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
|
<?xml version="1.0" encoding="UTF-8" standalone="no"?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title xmlns="">Howto: Porting newlib</title><link rel="stylesheet" href="./book_style.css" type="text/css" /><meta xmlns="" name="generator" content="DocBook XSL Stylesheets V1.75.2" /><link xmlns="" rel="shortcut icon" href="images/emb-16x16.png" type="image/png" /><link rel="home" href="#id2682085" title="Howto: Porting newlib" /><link rel="next" href="#id2699514" title="Chapter 1. Introduction" /><link rel="copyright" href="ln-id2699972.html" title="Legal Notice" /></head><body><div xmlns="" class="emblogoheader"><table width="100%" summary="Navigation header"><tr><th class="emblogo" align="left"><a href="http://www.embecosm.com"><img class="embnoborder" src="./images/embecosm.png" alt="Embecosm logo" width="154px" height="66px" /></a></th><th class="embstrapline" align="right" valign="bottom">
Services and Modeling for Embedded Software Development
</th></tr></table></div><div xmlns="" class="embrule"><img src="./images/strip-1024x1.gif" alt="Embecosm divider strip" width="1024px" height="10px" /></div><div xmlns="" class="embnavheader"><table width="100%" summary="Navigation header"><tr><th colspan="3" align="center" valign="bottom"></th></tr><tr></tr></table></div><div xml:lang="en_GB" class="book" title="Howto: Porting newlib" lang="en_GB"><div xmlns="" class="titlepage"><div><div><h1 xmlns="http://www.w3.org/1999/xhtml" class="title"><a id="id2682085"></a>
Howto: Porting <code class="systemitem">newlib</code>
</h1></div><div><h2 xmlns="http://www.w3.org/1999/xhtml" class="subtitle">
A Simple Guide
</h2></div><div><div xmlns="http://www.w3.org/1999/xhtml" class="author"><h3 class="author"><span class="firstname">Jeremy</span> <span class="surname">Bennett</span></h3></div></div><div><h3 xmlns="http://www.w3.org/1999/xhtml" class="corpauthor">
<a class="ulink" href="http://www.embecosm.com" target="_top">Embecosm</a>
</h3></div><div><p xmlns="http://www.w3.org/1999/xhtml" class="releaseinfo">
Application Note 9. Issue 1
</p></div><div><p xmlns="http://www.w3.org/1999/xhtml" class="pubdate">
July 2010
</p></div><div><p xmlns="http://www.w3.org/1999/xhtml" class="copyright">Copyright ©
2010
Embecosm Limited
</p></div><div><table class="legalnotice_link"><tr><td class="legalnotice_link"><a rel="license" href="http://creativecommons.org/licenses/by/2.0/uk/"><img alt="Creative Commons License" style="border-width:0" src="http://i.creativecommons.org/l/by/2.0/uk/88x31.png" /></a></td><td><span xmlns:dc="http://purl.org/dc/elements/1.1/" href="http://purl.org/dc/dcmitype/Text" property="dc:title" rel="dc:type">The document entitled
"
Howto: Porting newlib
"</span> by Jeremy Bennett of
<a xmlns:cc="http://creativecommons.org/ns#" href="http://www.embecosm.com/download.html" property="cc:attributionName" rel="cc:attributionURL">Embecosm</a> is licensed under a
<a rel="license" href="http://creativecommons.org/licenses/by/2.0/uk/">Creative
Commons Attribution 2.0 UK: England & Wales License</a>. See
the <a href="legalnotice.html">Legal Notice</a>
for details.
</td></tr></table></div></div><hr /></div><div class="toc"><p><b>Table of Contents</b></p><dl><dt><span class="chapter"><a href="#id2699514">1.
Introduction
</a></span></dt><dd><dl><dt><span class="sect1"><a href="#id2702274">1.1.
Target Audience
</a></span></dt><dt><span class="sect1"><a href="#id2706924">1.2.
Examples
</a></span></dt><dt><span class="sect1"><a href="#id2710284">1.3.
Further information
</a></span></dt><dt><span class="sect1"><a href="#id2700125">1.4.
About Embecosm Application Notes
</a></span></dt></dl></dd><dt><span class="chapter"><a href="#id2713251">2.
<code class="systemitem">newlib</code> within the <acronym class="acronym">GNU</acronym> Tool Chain
</a></span></dt><dd><dl><dt><span class="sect1"><a href="#sec_unified_source">2.1.
The Unified Source Tree
</a></span></dt><dd><dl><dt><span class="sect2"><a href="#id2714381">2.1.1.
Incorporating <code class="systemitem">Newlib</code> within the Tool Chain Build
</a></span></dt></dl></dd></dl></dd><dt><span class="chapter"><a href="#chap_overview">3.
Overview of <code class="systemitem">newlib</code>
</a></span></dt><dd><dl><dt><span class="sect1"><a href="#id2711887">3.1.
The relationship between <code class="systemitem">libgloss</code> and <code class="systemitem">newlib</code>
</a></span></dt><dt><span class="sect1"><a href="#sec_namespace_reent">3.2.
The C Namespace and Reentrant Functions
</a></span></dt><dt><span class="sect1"><a href="#sec_adding_target">3.3.
Adding a new Target to <code class="systemitem">Newlib</code>
</a></span></dt><dd><dl><dt><span class="sect2"><a href="#sec_configure_host">3.3.1.
Extending <code class="filename">configure.host</code> for a New Target
</a></span></dt></dl></dd></dl></dd><dt><span class="chapter"><a href="#chap_newlib">4.
Modifying <code class="systemitem">newlib</code>
</a></span></dt><dd><dl><dt><span class="sect1"><a href="#sec_machine_dir">4.1.
The Machine Directory
</a></span></dt><dd><dl><dt><span class="sect2"><a href="#sec_machine_dir_reconf">4.1.1.
Updating the Main Machine Directory Configuration files
</a></span></dt><dt><span class="sect2"><a href="#sec_setjmp_longjmp">4.1.2.
Implementing the <code class="function">setjmp</code> and
<code class="function">longjmp</code> functions.
</a></span></dt><dt><span class="sect2"><a href="#sec_machine_target_dir_reconf">4.1.3.
Updating the Target Specific Machine Directory Configuration files
</a></span></dt></dl></dd><dt><span class="sect1"><a href="#sec_newlib_headers">4.2.
Changing Headers
</a></span></dt><dd><dl><dt><span class="sect2"><a href="#sec_fp_header">4.2.1.
<acronym class="acronym">IEEE</acronym> Floating Point
</a></span></dt><dt><span class="sect2"><a href="#sec_setjmp_header">4.2.2.
<code class="function">setjmp</code> Buffer Size
</a></span></dt><dt><span class="sect2"><a href="#sec_newlib_config_header">4.2.3.
Miscellaneous System Definitions
</a></span></dt><dt><span class="sect2"><a href="#sec_other_headers">4.2.4.
Overriding Other Header Files
</a></span></dt></dl></dd></dl></dd><dt><span class="chapter"><a href="#chap_libgloss">5.
Modifying <code class="systemitem">libgloss</code>
</a></span></dt><dd><dl><dt><span class="sect1"><a href="#sec_platform_dir">5.1.
The Platform Directory
</a></span></dt><dd><dl><dt><span class="sect2"><a href="#sec_platform_dir_config">5.1.1.
Ensuring the Platform Directory is Configured
</a></span></dt></dl></dd><dt><span class="sect1"><a href="#sec_crt0">5.2.
The C Runtime Initialization, <code class="filename">crt0.o</code>
</a></span></dt><dd><dl><dt><span class="sect2"><a href="#sec_vectors">5.2.1.
Exception vector setup
</a></span></dt><dt><span class="sect2"><a href="#sec_stack_init">5.2.2.
The <code class="function">_start</code> Function and Stack Initialization
</a></span></dt><dt><span class="sect2"><a href="#id2717901">5.2.3.
Cache Initialization
</a></span></dt><dt><span class="sect2"><a href="#id2717944">5.2.4.
Clearing <acronym class="acronym">BSS</acronym>
</a></span></dt><dt><span class="sect2"><a href="#id2717990">5.2.5.
Constructor and Destructor Handling
</a></span></dt><dt><span class="sect2"><a href="#id2718120">5.2.6.
C Initialization Functions
</a></span></dt><dt><span class="sect2"><a href="#id2718170">5.2.7.
Invoking the main program
</a></span></dt></dl></dd><dt><span class="sect1"><a href="#sec_syscalls">5.3.
Standard System Call Implementations
</a></span></dt><dd><dl><dt><span class="sect2"><a href="#id2718360">5.3.1.
Error Handling
</a></span></dt><dt><span class="sect2"><a href="#id2718422">5.3.2.
The Global Environment, <code class="varname">environ</code>
</a></span></dt><dt><span class="sect2"><a href="#sec_exit">5.3.3.
Exit a program, <code class="function">_exit</code>
</a></span></dt><dt><span class="sect2"><a href="#id2718572">5.3.4.
Closing a file, <code class="function">close</code>
</a></span></dt><dt><span class="sect2"><a href="#id2718631">5.3.5.
Transfer Control to a New Process, <code class="function">execve</code>
</a></span></dt><dt><span class="sect2"><a href="#id2718701">5.3.6.
Create a new process, <code class="function">fork</code>
</a></span></dt><dt><span class="sect2"><a href="#id2718768">5.3.7.
Provide the Status of an Open File, <code class="function">fstat</code>
</a></span></dt><dt><span class="sect2"><a href="#id2718888">5.3.8.
Get the Current Process ID, <code class="function">getpid</code>
</a></span></dt><dt><span class="sect2"><a href="#id2718937">5.3.9.
Determine the Nature of a Stream, <code class="function">isatty</code>
</a></span></dt><dt><span class="sect2"><a href="#id2719051">5.3.10.
Send a Signal, <code class="function">kill</code>
</a></span></dt><dt><span class="sect2"><a href="#id2719106">5.3.11.
Rename an existing file, <code class="function">link</code>
</a></span></dt><dt><span class="sect2"><a href="#id2719160">5.3.12.
Set Position in a File, <code class="function">lseek</code>
</a></span></dt><dt><span class="sect2"><a href="#id2719266">5.3.13.
Open a file, <code class="function">open</code>
</a></span></dt><dt><span class="sect2"><a href="#id2719321">5.3.14.
Read from a File, <code class="function">read</code>
</a></span></dt><dt><span class="sect2"><a href="#sec_sbrk">5.3.15.
Allocate more Heap, <code class="function">sbrk</code>
</a></span></dt><dt><span class="sect2"><a href="#id2719729">5.3.16.
Status of a File (by Name), <code class="function">stat</code>
</a></span></dt><dt><span class="sect2"><a href="#id2719810">5.3.17.
Provide Process Timing Information, <code class="function">times</code>
</a></span></dt><dt><span class="sect2"><a href="#id2719864">5.3.18.
Remove a File's Directory Entry, <code class="function">unlink</code>
</a></span></dt><dt><span class="sect2"><a href="#id2719918">5.3.19.
Wait for a Child Process, <code class="function">wait</code>
</a></span></dt><dt><span class="sect2"><a href="#id2719973">5.3.20.
Write to a File, <code class="function">write</code>
</a></span></dt></dl></dd><dt><span class="sect1"><a href="#sec_reentrant_syscalls">5.4.
Reentrant System Call Implementations
</a></span></dt><dt><span class="sect1"><a href="#sec_bsp_config_make">5.5.
<acronym class="acronym">BSP</acronym> Configuration and <span class="command"><strong>Make</strong></span> file;
</a></span></dt><dd><dl><dt><span class="sect2"><a href="#id2720649">5.5.1.
<code class="filename">configure.in</code> for the <acronym class="acronym">BSP</acronym>
</a></span></dt><dt><span class="sect2"><a href="#id2720960">5.5.2.
<code class="filename">Makefile.in</code> for the <acronym class="acronym">BSP</acronym>
</a></span></dt></dl></dd><dt><span class="sect1"><a href="#sec_libnosys">5.6.
The Default <acronym class="acronym">BSP</acronym>, <code class="systemitem">libnosys</code>
</a></span></dt></dl></dd><dt><span class="chapter"><a href="#chap_build_install">6.
Configuring, Building and Installing <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</a></span></dt><dd><dl><dt><span class="sect1"><a href="#id2721527">6.1.
Configuring <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</a></span></dt><dt><span class="sect1"><a href="#id2721600">6.2.
Building <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</a></span></dt><dt><span class="sect1"><a href="#id2721651">6.3.
Testing <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</a></span></dt><dt><span class="sect1"><a href="#id2721699">6.4.
Installing <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</a></span></dt></dl></dd><dt><span class="chapter"><a href="#id2721746">7.
Modifying the <acronym class="acronym">GNU</acronym> Tool Chain
</a></span></dt><dd><dl><dt><span class="sect1"><a href="#sec_custom_newlib_loc">7.1.
Putting <code class="systemitem">Newlib</code> in a Custom Location
</a></span></dt><dt><span class="sect1"><a href="#sec_changing_gcc">7.2.
Changes to <acronym class="acronym">GCC</acronym>
</a></span></dt><dd><dl><dt><span class="sect2"><a href="#sec_gcc_machine_opts">7.2.1.
Adding Machine Specific Options for <code class="systemitem">Newlib</code>
</a></span></dt><dt><span class="sect2"><a href="#sec_gcc_specs">7.2.2.
Updating Spec Definitions
</a></span></dt></dl></dd><dt><span class="sect1"><a href="#sec_linker">7.3.
Changes to the <acronym class="acronym">GNU</acronym> Linker
</a></span></dt></dl></dd><dt><span class="chapter"><a href="#chap_testing">8.
Testing <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</a></span></dt><dd><dl><dt><span class="sect1"><a href="#sec_testing_newlib">8.1.
Testing <code class="systemitem">Newlib</code>
</a></span></dt><dd><dl><dt><span class="sect2"><a href="#id2723637">8.1.1.
Checking Physical Hardware
</a></span></dt></dl></dd><dt><span class="sect1"><a href="#sec_testing_libgloss">8.2.
Testing <code class="systemitem">Libgloss</code>
</a></span></dt></dl></dd><dt><span class="chapter"><a href="#chap_checklist">9.
Summary Checklist
</a></span></dt><dt><span class="glossary"><a href="#id2724638">
Glossary
</a></span></dt><dt><span class="bibliography"><a href="#id2724819">
References
</a></span></dt></dl></div><div class="chapter" title="Chapter 1. Introduction"><div class="titlepage"><div><div><h2 class="title"><a id="id2699514"></a>Chapter 1.
Introduction
</h2></div></div></div><p>
<code class="systemitem">Newlib</code> is a C library intended for use on embedded systems. It is a
conglomeration of several library parts, all under free software
licenses that make them easily usable on embedded products.
</p><div class="sect1" title="1.1. Target Audience"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="id2702274"></a>1.1.
Target Audience
</h2></div></div></div><p>
Porting <code class="systemitem">newlib</code> is not difficult, but advice for the beginner is thin
on the ground. This application note is intended for software
engineers porting <code class="systemitem">newlib</code> for the first time.
</p><p>
The detail of all the steps needed are covered here, and have been
tested using <code class="systemitem">newlib</code> versions 1.17.0 and 1.18.0 with the
<span class="application">OpenRISC 1000</span> .
</p><p>
For those who already have some experience, the entire porting process
is summarized in the final chapter, with links back to the main
document (see <a class="xref" href="#chap_checklist" title="Chapter 9. Summary Checklist">Chapter 9</a>). It's a
useful checklist when carrying out a new port.
</p></div><div class="sect1" title="1.2. Examples"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="id2706924"></a>1.2.
Examples
</h2></div></div></div><p>
This application note includes examples from the port of <code class="systemitem">newlib</code> to
the <span class="application">OpenRISC 1000</span> architecture, originally by Chris Bower, then of
Imperial College, London, and subsequently extensively updated by
Jeremy Bennett of Embecosm.
</p><p>
The examples are two <a class="firstterm" href="#id2724681"><em class="firstterm">Board Support Package</em></a>s
(<acronym class="acronym">BSP</acronym>) for use with the <span class="application">OpenRISC 1000</span> architectural simulator <span class="application">Or1ksim</span>
by the same two authors.
</p><p>
At the time of writing the <span class="application">OpenRISC 1000</span> implementation is
not part of the main <code class="systemitem">newlib</code> distribution. It can be downloaded from
OpenCores (<a class="ulink" href="http://www.opencores.org" target="_top">www.opencores.org</a>).
</p></div><div class="sect1" title="1.3. Further information"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="id2710284"></a>1.3.
Further information
</h2></div></div></div><p>
The main source of information is the <code class="systemitem">newlib</code> website (<a class="ulink" href="http://sourceware.org/newlib/" target="_top">sourceware.org/newlib</a>). This
includes a FAQ, which has brief instructions on porting <code class="systemitem">newlib</code> and
documentation for <code class="systemitem">libc</code> <a class="xref" href="#ref_libc" title="The Red Hat Newlib C Library">[1]</a> and <code class="systemitem">libm</code>, the
two libraries making up <code class="systemitem">newlib</code>. The <code class="systemitem">libc</code> documentation is
particularly useful, because it lists the system calls which must be
implemented by any new port, including minimal implementations.
</p><p>
The <code class="systemitem">newlib</code> <code class="filename">README</code> is another source of
information. Key header files within the source also contain useful
commenting, notably <code class="filename">ieeefp.h</code> and
<code class="filename">reent.h</code>.
</p><p>
There is also a mailing list, <code class="email"><<a class="email" href="mailto:newlib@sourceware.org"><a class="ulink" href="mailto:newlib@sourceware.org" target="_top">newlib@sourceware.org</a></a>></code>
where questions can be asked, or new ports submitted.
</p><p>
This application note does not cover the detail of testing <code class="systemitem">newlib</code>
on physical hardware. That subject is well covered by Dan Kegel's
<span class="emphasis"><em>Crosstool</em></span> project <a class="xref" href="#ref_crosstool" title="The Crosstool Project,">[2]</a>.
</p><p>
This application note has drawn heavily on these sources, and the
author would like to thank the providers of that original information.
</p></div><div class="sect1" title="1.4. About Embecosm Application Notes"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="id2700125"></a>1.4.
About Embecosm Application Notes
</h2></div></div></div><p>
Embecosm is a consultancy specializing in hardware
modeling and open source tool chains for the embedded market. If we
can ever be of help, please get in touch at
<code class="email"><<a class="email" href="mailto:sales@embecosm.com">sales@embecosm.com</a>></code>.
</p><p>
As part of its commitment to the open source community, Embecosm
publishes a series of free and open source application notes, designed
to help working engineers with practical problems.
</p><p>
Feedback is always welcome, which should be sent to
<code class="email"><<a class="email" href="mailto:info@embecosm.com">info@embecosm.com</a>></code>.
</p></div></div><div class="chapter" title="Chapter 2. newlib within the GNU Tool Chain"><div class="titlepage"><div><div><h2 class="title"><a id="id2713251"></a>Chapter 2.
<code class="systemitem">newlib</code> within the <acronym class="acronym">GNU</acronym> Tool Chain
</h2></div></div></div><p>
<code class="systemitem">Newlib</code> is intended for use with the GNU tool chain. If <code class="systemitem">newlib</code> is
included within the build of the GNU tool chain, then all the
libraries will be built and installed in the correct places to be found
by <acronym class="acronym">GCC</acronym>
</p><div class="sect1" title="2.1. The Unified Source Tree"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_unified_source"></a>2.1.
The Unified Source Tree
</h2></div></div></div><p>
The three separate packages, <code class="systemitem">binutils</code>, <acronym class="acronym">GCC</acronym> and <acronym class="acronym">GDB</acronym> are all taken
from a common source tree. <acronym class="acronym">GCC</acronym> and <acronym class="acronym">GDB</acronym> both use many libraries
from <code class="systemitem">binutils</code>. It is convenient to reassemble that source tree and
make a single build of all the tools together.
</p><p>
The easiest way to achieve this is to link all the top level
directories in each package into a single unified directory, leaving
out any duplicated files or directories.
</p><p>
The following <span class="command"><strong>bash</strong></span> script will take unpacked distributions of
<code class="systemitem">binutils</code> <acronym class="acronym">GCC</acronym> and <acronym class="acronym">GDB</acronym> and link them into a single directory,
<code class="filename">srcw</code>.
</p><div class="informalfigure"><pre class="programlisting">
#!/bin/bash
component_dirs='binutils-2.18.50 gcc-4.2.2 gdb-6.8'
unified_src=srcw
cd ${unified_src}
ignore_list=". .. CVS .svn"
for srcdir in ${component_dirs}
do
echo "Component: $srcdir"
case srcdir
in
/* | [A-Za-z]:[\\/]*)
;;
*)
srcdir="../${srcdir}"
;;
esac
files=`ls -a ${srcdir}`
for f in ${files}
do
found=
for i in ${ignore_list}
do
if [ "$f" = "$i" ]
then
found=yes
fi
done
if [ -z "${found}" ]
then
echo "$f ..linked"
ln -s ${srcdir}/$f .
fi
done
ignore_list="${ignore_list} ${files}"
done
cd ..
</pre></div><p>
The entire tool chain can then be configured and built in a separate
directory. The <code class="filename">configure</code> script understands to
pass on top level arguments to subsidiary configuration scripts. For
example to configure to build a C only tool chain for the 32-bit
<span class="application">OpenRISC 1000</span> architecture to be installed in
<code class="filename">/opt/or32-elf</code>, the following would be
appropriate.
</p><div class="informalfigure"><pre class="programlisting">
mkdir build
cd build
../src/configure --target=or32-elf --enable-languages=c --prefix=/opt/or32-elf
cd ..
</pre></div><p>
Each tool can be built with its own specific target within that build
directory
</p><div class="informalfigure"><pre class="programlisting">
cd build
make all-build all-binutils all-gas all-ld all-gcc all-gdb
cd ..
</pre></div><div class="note" title="Note" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Note"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Note]" src="./images/note.png" /></td><th align="left">Note</th></tr><tr><td align="left" valign="top"><p>
The initial <span class="command"><strong>make</strong></span> target, <code class="literal">all-build</code> is used to
build some of the baseline libraries and tools used throughout the
tool chain.
</p></td></tr></table></div><p>
Similarly the tools can be installed using the following:
</p><div class="informalfigure"><pre class="programlisting">
cd build
make install-build install-binutils install-gas install-ld install-gcc \
install-gdb
cd ..
</pre></div><div class="sect2" title="2.1.1. Incorporating Newlib within the Tool Chain Build"><div class="titlepage"><div><div><h3 class="title"><a id="id2714381"></a>2.1.1.
Incorporating <code class="systemitem">Newlib</code> within the Tool Chain Build
</h3></div></div></div><p>
<code class="systemitem">Newlib</code> can be linked into the unified source directory in the same
fashion. All that is needed is to add <code class="systemitem">newlib</code> to the component
directories in the linking script.
</p><div class="informalfigure"><pre class="programlisting">
#!/bin/bash
component_dirs='binutils-2.18.50 gcc-4.2.2 newlib-1.18.0 gdb-6.8'
unified_src=srcw
...
</pre></div><p>
The configuration command should also specify that this is a build
using <code class="systemitem">newlib</code>
</p><div class="informalfigure"><pre class="programlisting">
mkdir build
cd build
../src/configure --target=or32-elf --enable-languages=c --with-newlib \
--prefix=/opt/or32-elf
cd ..
</pre></div><p>
Two new targets are needed for <code class="systemitem">newlib</code>, one to build <code class="systemitem">newlib</code> itself,
and one to build any board support packages using <code class="systemitem">libgloss</code> (see
<a class="xref" href="#chap_overview" title="Chapter 3. Overview of newlib">Chapter 3</a> for an explanation of how <code class="systemitem">libgloss</code>
is used with <code class="systemitem">newlib</code>).
</p><div class="informalfigure"><pre class="programlisting">
cd build
make all-build all-binutils all-gas all-ld all-gcc all-target-newlib \
all-target-libgloss all-gdb
cd ..
</pre></div><p>
Similarly additional targets are needed for installation.
</p><div class="informalfigure"><pre class="programlisting">
cd build
make install-build install-binutils install-gas install-ld install-gcc \
install-target-newlib install-target-libgloss install-gdb
cd ..
</pre></div></div></div></div><div class="chapter" title="Chapter 3. Overview of newlib"><div class="titlepage"><div><div><h2 class="title"><a id="chap_overview"></a>Chapter 3.
Overview of <code class="systemitem">newlib</code>
</h2></div></div></div><p>
</p><div class="sect1" title="3.1. The relationship between libgloss and newlib"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="id2711887"></a>3.1.
The relationship between <code class="systemitem">libgloss</code> and <code class="systemitem">newlib</code>
</h2></div></div></div><p>
<code class="systemitem">Newlib</code> is now divided into two parts. The main
<code class="filename">newlib</code> directory contains the bulk of the code
for the two main libraries, <code class="systemitem">libc</code> and <code class="systemitem">libm</code>, together with any
<span class="emphasis"><em>architecture</em></span> specific code for particular
targets.
</p><p>
The <code class="filename">libgloss</code> directory contains code specific to
particular platforms on which the library will be used, generally
referred to as the <a class="firstterm" href="#id2724681"><em class="firstterm">Board Support Package</em></a>
(<acronym class="acronym">BSP</acronym>). Any particular target architecture may have multiple <acronym class="acronym">BSP</acronym>s,
for example for different hardware platforms, for a simulator etc.
</p><p>
The target architecture specific code within the
<code class="filename">newlib</code> directory may be very modest - possibly as
little as an implementation of <code class="function">setjmp</code> and a
specification of the <acronym class="acronym">IEEE</acronym> floating point format to use.
</p><p>
The board support package is more complex. It requires an
implementation of eighteen system calls and the definition of one
global data structure, although the implementation of some of those
system calls may be completely trivial.
</p><div class="note" title="Note" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Note"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Note]" src="./images/note.png" /></td><th align="left">Note</th></tr><tr><td align="left" valign="top"><p>
The separation of <acronym class="acronym">BSP</acronym> implementation into <code class="systemitem">libgloss</code> is relatively
recent. Consequently the source tree contains a number of older
target implementations where the <acronym class="acronym">BSP</acronym> is entirely within
<code class="systemitem">newlib</code>. When looking for examples, be sure to choose an
architecture which has been implemented through <code class="systemitem">libgloss</code>. The
<span class="application">OpenRISC 1000</span> implementation is one such architecture.
</p></td></tr></table></div></div><div class="sect1" title="3.2. The C Namespace and Reentrant Functions"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_namespace_reent"></a>3.2.
The C Namespace and Reentrant Functions
</h2></div></div></div><p>
The <acronym class="acronym">BSP</acronym> implements the system calls—functions like
<code class="function">close</code>, <code class="function">write</code> etc. It is
possible for the <acronym class="acronym">BSP</acronym> to implement these directly, but these will
then be defined in the main C namespace. It is perfectly permissible
for the user to replace these functions, and the user versions take
precedence, which requires some care at link time.
</p><p>
<code class="systemitem">Newlib</code> allows the implementer instead to provide namespace clean
versions of these functions by prefixing them with an
underscore. <code class="systemitem">Newlib</code> will ensure that the system calls map to these
namespace clean version (i.e. a call to <code class="function">close</code>
becomes a call to <code class="function">_close</code>) unless the user has
reimplemented that function themselves.
</p><p>
A <a class="firstterm" href="#id2724738"><em class="firstterm">reentrant</em></a> function may be safely called from
a second thread, while a first thread of control is executing. In
general a function that modifies no static or global state, will be
reentrant.
</p><p>
Many system calls are trivially reentrant. However for some calls,
reentrancy is not easy to provide automatically, so reentrant versions
are provided. Thus for <code class="function">close</code>, there is the
reentrant version <code class="function">close_r</code>. The reentrant versions
take an extra argument, a <span class="emphasis"><em>reentrancy structure</em></span>,
which can be used to ensure correct behavior, by providing per-thread
versions of global data structures.
</p><p>
It is worth noting that use of the global error value,
<code class="varname">errno</code> is a common source of non-reentrancy. The
standard reentrancy structure includes an entry for a per-thread value
of <code class="varname">errno</code>.
</p><p>
For many systems, the issue of reentrancy does not arise. If there is
only ever one thread of control, or if separate threads have their own
address space there is no problem.
</p><p>
However it's worth remembering that even a bare metal system may
encounter issues with reentrancy if event handlers are allowed to use
the system calls.
</p><p>
<code class="systemitem">Newlib</code> gives considerable flexibility, particularly where namespace
clean versions of the basic system calls are implemented. The
implementer can choose to provide implementations of the reentrant
versions of the functions. Alternatively <code class="systemitem">newlib</code> can provide
reentrancy at the library level, but mapping the calls down the system
calls, which are not themselves reentrant. This last can often prove
a practical solution to the problem.
</p></div><div class="sect1" title="3.3. Adding a new Target to Newlib"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_adding_target"></a>3.3.
Adding a new Target to <code class="systemitem">Newlib</code>
</h2></div></div></div><p>
Adding a new architecture to <code class="systemitem">newlib</code> requires the following steps.
</p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>
Provide a machine specific directory within the
<code class="filename">newlib</code> directory for architecture specific
code, notably the <code class="function">setjmp</code> implementation.
</p></li><li class="listitem"><p>
Provide a platform directory for <acronym class="acronym">BSP</acronym> implementation(s) within
the <code class="filename">libgloss</code> directory. The code implementing
systems calls for each <acronym class="acronym">BSP</acronym> is placed in this directory.
</p></li><li class="listitem"><p>
Update the <code class="filename">configure.host</code> file in the
<code class="filename">newlib</code> directory to point to the machine and
platform directories for the new target.
</p></li></ol></div><div class="sect2" title="3.3.1. Extending configure.host for a New Target"><div class="titlepage"><div><div><h3 class="title"><a id="sec_configure_host"></a>3.3.1.
Extending <code class="filename">configure.host</code> for a New Target
</h3></div></div></div><p>
The <code class="filename">configure.host</code> file needs changes in two
places, to identify the architecture specific machine directory and
the platform directory for <acronym class="acronym">BSP</acronym> implementations.
</p><p>
The machine name is specified in a <code class="literal">case</code> switch on
the <code class="literal">${host_cpu}</code> early on in the file. Add a new
<code class="literal">case</code> entry defining
<code class="literal">machine_type</code> for the architecture. Thus for
<span class="application">OpenRISC 1000</span> 32-bit architecture we have:
</p><div class="informalfigure"><pre class="programlisting">
or32)
machine_dir=or32
;;
</pre></div><p>
This specifies that the machine specific code for this architecture
will be found in the directory
<code class="filename">newlib/libc/machine/or32</code>.
</p><p>
The platform directory and details are specified in a subsequent
<code class="literal">case</code> switch on <code class="literal">${host}</code>
(i.e. the full triplet, not just the <acronym class="acronym">CPU</acronym> type).
For the 32-bit <span class="application">OpenRISC 1000</span> we have the following.
</p><div class="informalfigure"><pre class="programlisting">
or32-*-*)
syscall_dir=syscalls
;;
</pre></div><p>
This is the simplest option, specifying that the <acronym class="acronym">BSP</acronym>s for all
<span class="application">OpenRISC 1000</span> 32-bit targets will implement namespace clean system calls,
and rely on <code class="systemitem">newlib</code> to map reentrant calls down to them. The
directory name for the <acronym class="acronym">BSP</acronym> implementations will match that of the
machine directory, but within the <code class="filename">libgloss</code>
directory. So for <span class="application">OpenRISC 1000</span> 32-bit targets; the <acronym class="acronym">BSP</acronym> implementations
are in <code class="filename">libgloss/or32</code>.
</p><p>
There are four common alternatives for specifying how the <acronym class="acronym">BSP</acronym>
will be implemented.
</p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>
The implementer defines reentrant namespace clean versions of
the system calls. In this case, <code class="literal">syscall_dir</code>
is set to <code class="literal">syscalls</code> as above, but in addition,
<code class="literal">-DREENTRANT_SYSCALLS_PROVIDED</code> is added to
<code class="literal">newlib_cflags</code> in
<code class="filename">configure.host</code>. For the <span class="application">OpenRISC 1000</span> 32-bit
target we could have done this with:
</p><div class="informalfigure"><pre class="programlisting">
or32-*-*)
syscall_dir=syscalls
newlib_cflags="${newlib_cflags} -DREENTRANT_SYSCALLS_PROVIDED"
;;
</pre></div><p>
</p><p>
For convenience, stub versions of the reentrant functions may
be found in the <code class="literal">libc/reent</code> directory. These
are in fact the functions used if the reentrant system calls are
not provided, and map to the non-reentrant versions.
</p></li><li class="listitem"><p>
The implementer defines non-reentrant, but namespace clean
versions of the system calls. This is the approach we have used
with the <span class="application">OpenRISC 1000</span> and all the implementer needs to do in this case
is to set <code class="literal">syscall_dir</code> to
<code class="literal">syscalls</code> in
<code class="filename">configure.host</code>. <code class="systemitem">newlib</code> will map reentrant
calls down to the non-reentrant versions.
</p></li><li class="listitem"><p>
The implementer defines non-reentrant, regular
versions of the system calls (i.e. <code class="function">close</code>
rather than <code class="function">_close</code>). The library will
be neither reentrant, not namespace clean, but will work. In
this case,
<code class="literal">-DMISSING_SYSCALL_NAMES</code> is
added to <code class="literal">newlib_cflags</code> in
<code class="filename">configure.host</code>. For the <span class="application">OpenRISC 1000</span> we could
have done this with:
</p><div class="informalfigure"><pre class="programlisting">
or32-*-*)
newlib_cflags="${newlib_cflags} -DMISSING_SYSCALL_NAMES"
;;
</pre></div><p>
</p><p>
Note in particular that <code class="literal">syscall_dir</code> is not
defined in this case.
</p></li><li class="listitem"><p>
The implementer defines non-reentrant, regular
versions of the system calls (i.e. <code class="function">close</code>
rather than <code class="function">_close</code>). The reentrant system
calls are mapped onto these functions. The library will
not be namespace clean, but will offer reentrancy at the library
level. In
this case,
<code class="literal">-DMISSING_SYSCALL_NAMES</code> and
<code class="literal">-DREENTRANT_SYSCALLS_PROVIDED</code> are both
added to <code class="literal">newlib_cflags</code> in
<code class="filename">configure.host</code>. For the <span class="application">OpenRISC 1000</span> we could
have done this with:
</p><div class="informalfigure"><pre class="programlisting">
or32-*-*)
newlib_cflags="${newlib_cflags} -DMISSING_SYSCALL_NAMES"
newlib_cflags="${newlib_cflags} -DREENTRANT_SYSCALLS_PROVIDED"
;;
</pre></div><p>
</p><p>
Note in particular that <code class="literal">syscall_dir</code> is not
defined in this case.
</p></li></ol></div></div></div></div><div class="chapter" title="Chapter 4. Modifying newlib"><div class="titlepage"><div><div><h2 class="title"><a id="chap_newlib"></a>Chapter 4.
Modifying <code class="systemitem">newlib</code>
</h2></div></div></div><p>
Changes that depend on the architecture, and not the particular platform
being used, are made in the <code class="filename">newlib</code> directory. These
comprise changes to standard headers and custom code for the
architecture.
</p><div class="sect1" title="4.1. The Machine Directory"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_machine_dir"></a>4.1.
The Machine Directory
</h2></div></div></div><p>
Within the <code class="filename">newlib</code> directory, machine specific
code is placed in a target specific directory,
<code class="filename">libc/machine/<em class="replaceable"><code>arch</code></em></code>.
</p><p>
The only code that has to be there is the implementation of
<code class="function">setjmp</code> and <code class="function">longjmp</code>, since
the implementation of these two functions invariably requires target
specific machine code. However any other target specific code may
also be placed here.
</p><div class="sect2" title="4.1.1. Updating the Main Machine Directory Configuration files"><div class="titlepage"><div><div><h3 class="title"><a id="sec_machine_dir_reconf"></a>4.1.1.
Updating the Main Machine Directory Configuration files
</h3></div></div></div><p>
The machine directory uses <acronym class="acronym">GNU</acronym> <span class="application">autoconf</span> and <span class="application">automake</span> for
configuration. There is a configuration template file
(<code class="filename">configure.in</code>) and Makefile template
(<code class="filename">Makefile.am</code>) in the main machine directory
(<code class="filename">libc/machine</code> within the
<code class="filename">newlib</code> directory).
</p><p>
<code class="filename">configure.ac</code> contains a <code class="literal">case</code>
statement configuring the target specific subdirectories. This must
be updated to configure the subdirectory for the new target. Thus
for the <span class="application">OpenRISC 1000</span> we have the following.
</p><div class="informalfigure"><pre class="programlisting">
if test -n "${machine_dir}"; then
case ${machine_dir} in
a29k) AC_CONFIG_SUBDIRS(a29k) ;;
arm) AC_CONFIG_SUBDIRS(arm) ;;
<other machines not shown>
necv70) AC_CONFIG_SUBDIRS(necv70) ;;
or32) AC_CONFIG_SUBDIRS(or32) ;;
powerpc) AC_CONFIG_SUBDIRS(powerpc) ;;
<other machines not shown>
xstormy16) AC_CONFIG_SUBDIRS(xstormy16) ;;
z8k) AC_CONFIG_SUBDIRS(z8k) ;;
esac;
fi
</pre></div><p>
<code class="filename">Makefile.am</code> is standard and will not need to be
changed. Having changed the configuration template, the
configuration file, <code class="filename">configure</code>, will need to be
regenerated. This only requires running <span class="application">autoconf</span>
</p><div class="informalfigure"><pre class="programlisting">
autoconf
</pre></div><p>
Since <code class="filename">Makefile.am</code> has not been changed there is
no need to run <span class="application">automake</span>
</p></div><div class="sect2" title="4.1.2. Implementing the setjmp and longjmp functions."><div class="titlepage"><div><div><h3 class="title"><a id="sec_setjmp_longjmp"></a>4.1.2.
Implementing the <code class="function">setjmp</code> and
<code class="function">longjmp</code> functions.
</h3></div></div></div><p>
<code class="function">setjmp</code> and <code class="function">longjmp</code> are a
pair of C function facilitating cross-procedure transfer of
control. Typically they are used to allow resumption of execution at
a known good point after an error.
</p><p>
Both take as first argument a buffer, which is used to hold the
machine state at the jump destination. When
<code class="function">setjmp</code> is called it populates that buffer with
the current location state (which includes stack and frame pointers
and the return address for the call to <code class="function">setjmp</code>,
and returns zero.
</p><p>
<code class="function">longjmp</code> takes a buffer previously populated by
<code class="function">setjmp</code>. It also takes a (non-zero) second
argument, which will ultimately be the result of the function
call. <code class="function">longjmp</code> restores the machine state from
the buffer. It then jumps to the return address it has just
restored, passing its second argument as the result. That return
address is the return address from the original call to
<code class="function">setjmp</code>, so the effect will be as if
<code class="function">setjmp</code> has just returned with a non-zero
argument.
</p><p>
<code class="function">setjmp</code> and <code class="function">longjmp</code> are
typically used in a top level function in the following way.
</p><div class="informalfigure"><pre class="programlisting">
#include <setjmp.h>
...
jmp_buf buf;
if (0 == setjmp (buf))
{
<span class="emphasis"><em>normal processing passing in buf</em></span>
}
else
{
<span class="emphasis"><em>error handling code</em></span>
}
...
</pre></div><p>
During normal processing if an error is found, the state held in
<code class="varname">buf</code> can be used to return control back to the top
level using <code class="function">longjmp</code>.
</p><div class="informalfigure"><pre class="programlisting">
#include <setjmp.h>
...
if (<span class="emphasis"><em>error detected</em></span>)
{
longjmp (buf, 1);
}
...
</pre></div><p>
The program will behave as though the original call to
<code class="function">setjmp</code> had just returned with result 1.
</p><p>
It will be appreciated that this is behavior that cannot usually be
written in C. The <span class="application">OpenRISC 1000</span> implementation is given as an example. This
processor has 32 registers, <code class="literal">r0</code> through
<code class="literal">r31</code>, each of 32-bits. <code class="literal">r0</code> is
always tied to zero, so need not be saved. <code class="literal">r11</code> is
the function result register, which is always set by
<code class="function">setjmp</code> and <code class="function">longjmp</code>, so
also need not be saved. In addition we should save and restore the
machine's 32-bit supervision register, which holds the branch flag.
</p><p>
Thus we need the buffer to be 31 32-bit words long. This is defined
in the <code class="function">setjmp</code> header (see <a class="xref" href="#sec_setjmp_header" title="4.2.2. setjmp Buffer Size">Section 4.2.2</a>).
</p><p>
In the <a class="firstterm" href="#id2724644"><em class="firstterm">Application Binary Interface</em></a> (<acronym class="acronym">ABI</acronym>)
for the <span class="application">OpenRISC 1000</span> , function arguments are passed in registers
<code class="literal">r3</code> through <code class="literal">r8</code> and the function
return address is in <code class="literal">r9</code>.
</p><p>
When defining these two functions, in assembler, be aware of any
prefix conventions used by the C compiler. It is common for symbols
defined in C to have an underscore prepended (this is the case for
the <span class="application">OpenRISC 1000</span> ). Thus in this case the assembler should define
<code class="function">_setjmp</code> and <code class="function">_longjmp</code>.
</p><p>
This is the implementation of <code class="function">setjmp</code>.
</p><div class="informalfigure"><pre class="programlisting">
.global _setjmp
_setjmp:
l.sw 4(r3),r1 /* Slot 0 saved for flag in future */
l.sw 8(r3),r2
l.sw 12(r3),r3
l.sw 16(r3),r4
l.sw 20(r3),r5
l.sw 24(r3),r6
l.sw 28(r3),r7
l.sw 32(r3),r8
l.sw 36(r3),r9
l.sw 40(r3),r10 /* Skip r11 */
l.sw 44(r3),r12
l.sw 48(r3),r13
l.sw 52(r3),r14
l.sw 56(r3),r15
l.sw 60(r3),r16
l.sw 64(r3),r17
l.sw 68(r3),r18
l.sw 72(r3),r19
l.sw 76(r3),r20
l.sw 80(r3),r21
l.sw 84(r3),r22
l.sw 88(r3),r23
l.sw 92(r3),r24
l.sw 96(r3),r25
l.sw 100(r3),r26
l.sw 104(r3),r27
l.sw 108(r3),r28
l.sw 112(r3),r29
l.sw 116(r3),r30
l.sw 120(r3),r31
l.jr r9
l.addi r11,r0,0 /* Zero result */
</pre></div><p>
In this simplified implementation, the status flags are not
saved—that is a potential future enhancement. All the general
registers, with the exception of <code class="literal">r0</code> (always zero)
and <code class="literal">r11</code> (result register) are saved in the
buffer, which, being the first argument, is pointed to by
<code class="literal">r3</code>.
</p><p>
Finally the result register, <code class="literal">r11</code> is set to zero
and the function returns using <code class="literal">r9</code> (the <span class="application">OpenRISC 1000</span> has
delayed branches, so the setting of <code class="literal">r11</code> is placed
after the branch to return.).
</p><p>
The implementation of <code class="function">longjmp</code> is slightly more
complex, since the second argument will be returned as the effective
result from <code class="function">setjmp</code>, <span class="emphasis"><em>unless</em></span>
the second argument is zero in which case 1 is used.
</p><p>
The result must be dealt with first and placed in the result
register, <code class="literal">r11</code>, because the second argument, in
<code class="literal">r4</code> will be subsequently overwritten when the
machine state is restored. Similarly we must ensure that
<code class="literal">r3</code>, which holds the first argument pointing to
the restore buffer must itself be the last register restored.
</p><div class="informalfigure"><pre class="programlisting">
.global _longjmp
_longjmp:
/* Sort out the return value */
l.sfne r4,r0
l.bf 1f
l.nop
l.j 2f
l.addi r11,r0,1 /* 1 as result */
1: l.addi r11,r4,0 /* val as result */
/* Restore all the other registers, leaving r3 to last. */
2: l.lwz r31,120(r3)
l.lwz r30,116(r3)
l.lwz r29,112(r3)
l.lwz r28,108(r3)
l.lwz r27,104(r3)
l.lwz r26,100(r3)
l.lwz r25,96(r3)
l.lwz r24,92(r3)
l.lwz r23,88(r3)
l.lwz r22,84(r3)
l.lwz r21,80(r3)
l.lwz r20,76(r3)
l.lwz r19,72(r3)
l.lwz r18,68(r3)
l.lwz r17,64(r3)
l.lwz r16,60(r3)
l.lwz r15,56(r3)
l.lwz r14,52(r3)
l.lwz r13,48(r3)
l.lwz r12,44(r3)
l.lwz r10,40(r3) /* Omit r11 */
l.lwz r9,36(r3)
l.lwz r8,32(r3)
l.lwz r7,28(r3)
l.lwz r6,24(r3)
l.lwz r5,20(r3)
l.lwz r4,16(r3)
l.lwz r2,8(r3) /* Skip r3 */
l.lwz r1,4(r3) /* Slot 0 saved for flag in future */
l.lwz r3,12(r3) /* Now safe */
/* Result is already in r11. Having restored r9, it will appear as
though we have returned from the earlier call to _setjmp. The
non-zero result gives it away though. */
l.jr r9
l.nop
</pre></div><p>
The return address, stack pointer and frame pointer having been
restored, the return from the function, will place the execution
point immediately after the original call to
<code class="function">setjmp</code>.
</p><p>
The following is a simple test program, which can be used to verify
that <code class="function">setjmp</code> and <code class="function">longjmp</code>
are working correctly.
</p><div class="informalfigure"><pre class="programlisting">
#include <setjmp.h>
#include <stdio.h>
void
testit (jmp_buf env,
int prev_res)
{
int res = (0 == prev_res) ? prev_res : prev_res + 1;
printf ("Long jumping with result %d\n", res);
longjmp (env, res);
} /* testit () */
int
main (int argc,
char *argv[])
{
jmp_buf env;
int res = setjmp (env);
printf ("res = 0x%08x\n", res);
if (res > 1)
{
return 0;
}
testit (env, res);
return 256; /* We should never actually get here */
} /* main () */
</pre></div></div><div class="sect2" title="4.1.3. Updating the Target Specific Machine Directory Configuration files"><div class="titlepage"><div><div><h3 class="title"><a id="sec_machine_target_dir_reconf"></a>4.1.3.
Updating the Target Specific Machine Directory Configuration files
</h3></div></div></div><p>
<code class="filename">configure.in</code> and
<code class="filename">Makefile.am</code> files also be needed for the target
specific directory (i.e.
<code class="filename">libc/machine/<em class="replaceable"><code>target</code></em></code>
within the <code class="filename">newlib</code> directory). These are
generally quite standard, and the easiest approach is to copy the
versions used for the <span class="application">fr30</span> architecture. Modern practice is to use
the file name <code class="filename">configure.ac</code> rather than
<code class="filename">configure.in</code>, but either will be accepted by
<span class="application">autoconf</span>.
</p><p>
<code class="filename">Makefile.am</code> should be modified if necessary to
specify the source files (for example <code class="filename">setjmp.S</code>
and <code class="filename">longjmp.S</code>). More complex implementations
may require modifications to <code class="filename">configure.in</code> as
well.
</p><p>
For example the <span class="application">OpenRISC 1000</span> machine directory
(<code class="filename">libc/machine/or32</code> within the
<code class="filename">newlib</code> directory) contains the following.
</p><div class="informalfigure"><pre class="programlisting">
AUTOMAKE_OPTIONS = cygnus
INCLUDES = $(NEWLIB_CFLAGS) $(CROSS_CFLAGS) $(TARGET_CFLAGS)
AM_CCASFLAGS = $(INCLUDES)
noinst_LIBRARIES = lib.a
lib_a_SOURCES = longjmp.S setjmp.S
lib_a_CCASFLAGS=$(AM_CCASFLAGS)
lib_a_CFLAGS=$(AM_CFLAGS)
ACLOCAL_AMFLAGS = -I ../../.. -I ../../../..
CONFIG_STATUS_DEPENDENCIES = $(newlib_basedir)/configure.host
</pre></div><p>
After any changes it will be necessary to run <span class="application">autoconf</span> and/or
<span class="application">automake</span>
to generate new versions of <code class="filename">configure</code> and
<code class="filename">Makefile.in</code>. <span class="application">autoconf</span> requires a number of
<code class="systemitem">newlib</code> specific macros. These can be generated from the main
<code class="systemitem">newlib</code> include file (<code class="filename">acinclude.m4</code>) by running
<span class="application">aclocal</span>. The full set of commands would be.
</p><div class="informalfigure"><pre class="programlisting">
aclocal -I ../../..
autoconf
automake --cygnus Makefile
</pre></div><p>
<span class="application">aclocal</span> only need to be run the first time the directory is
created, or when moving the directory to a new release of
<code class="systemitem">newlib</code>. <span class="application">autoconf</span> need only be run each time
<code class="filename">configure.in</code> (or
<code class="filename">configure.ac</code>) is changed. <span class="application">automake</span> need only
be run each time <code class="filename">Makefile.am</code> is changed.
</p></div></div><div class="sect1" title="4.2. Changing Headers"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_newlib_headers"></a>4.2.
Changing Headers
</h2></div></div></div><p>
There are two places, where header definitions must be modified for a
new target architecture: the specification of the <acronym class="acronym">IEEE</acronym> floating
point format used, and the specification of the
<code class="function">setjmp</code> buffer size.
</p><div class="sect2" title="4.2.1. IEEE Floating Point"><div class="titlepage"><div><div><h3 class="title"><a id="sec_fp_header"></a>4.2.1.
<acronym class="acronym">IEEE</acronym> Floating Point
</h3></div></div></div><p>
The floating point format is specified within the
<code class="filename">newlib</code> directory in
<code class="filename">libc/include/machine/ieeefp.h</code>. Details of how
the <acronym class="acronym">IEEE</acronym> 754 format is implemented, and variations from the
standard, are specified by defining a number of C macros.
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
<code class="literal">__IEEE_BIG_ENDIAN</code>
</p><p>
Define this macro if the floating point format is
<a class="firstterm" href="#id2724662"><em class="firstterm">big endian</em></a>.
</p><div class="caution" title="Caution" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Caution"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Caution]" src="./images/caution.png" /></td><th align="left">Caution</th></tr><tr><td align="left" valign="top"><p>
One, and only one of <code class="literal">__IEEE_BIG_ENDIAN</code>
and <code class="literal">__IEEE_LITTLE_ENDIAN</code> must be defined.
</p></td></tr></table></div></li><li class="listitem"><p>
<code class="literal">__IEEE_LITTLE_ENDIAN</code>
</p><p>
Define this macro if the floating point format is
<a class="firstterm" href="#id2724720"><em class="firstterm">little endian</em></a>.
</p><div class="caution" title="Caution" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Caution"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Caution]" src="./images/caution.png" /></td><th align="left">Caution</th></tr><tr><td align="left" valign="top"><p>
One, and only one of <code class="literal">__IEEE_LITTLE_ENDIAN</code>
and <code class="literal">__IEEE_BIG_ENDIAN</code> must be defined.
</p></td></tr></table></div></li><li class="listitem"><p>
<code class="literal">__IEEE_BYTES_LITTLE_ENDIAN</code>
</p><p>
Define this macro in addition to
<code class="literal">__IEEE_BIG_ENDIAN</code>, where the words of a
multi-word <acronym class="acronym">IEEE</acronym> floating point number are in big endian
order, but the bytes within each word are in little endian
order.
</p></li><li class="listitem"><p>
<code class="literal">_DOUBLE_IS_32BITS </code>
</p><p>
Define this if double precision floating point is represented
using the 32-bit <acronym class="acronym">IEEE</acronym> representation.
</p></li><li class="listitem"><p>
<code class="literal">_FLOAT_ARG</code>
</p><p>
Floating point arguments are usually promoted to double when
passed as arguments. If this is not the case, then this macro
should be defined to the type actually used to pass floating
point arguments.
</p></li><li class="listitem"><p>
<code class="literal">_FLT_LARGEST_EXPONENT_IS_NORMAL</code>
</p><p>
Define this if the floating point format uses the largest
exponent for finite numbers rather than <acronym class="acronym">NaN</acronym> and
infinities. Such a format cannot represent <acronym class="acronym">NaN</acronym>s or infinities,
but it's <code class="literal">FLT_MAX</code> is twice the standard <acronym class="acronym">IEEE</acronym>
value.
</p></li><li class="listitem"><p>
<code class="literal">_FLT_NO_DENORMALS</code>
</p><p>
Define this if the floating point format does not support <acronym class="acronym">IEEE</acronym>
denormalized numbers. In this case, every floating point number
with a zero exponent is treated as a zero representation.
</p></li></ul></div><div class="caution" title="Caution" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Caution"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Caution]" src="./images/caution.png" /></td><th align="left">Caution</th></tr><tr><td align="left" valign="top"><p>
Two of these macros
(<code class="literal">_FLT_LARGEST_EXPONENT_IS_NORMAL</code> and
<code class="literal">_FLT_NO_DENORMALS</code>) specify deviations from
<acronym class="acronym">IEEE</acronym> 754. These macros only work with single-precision floating
point and may not work correctly if hardware floating point
support is used (enabled by configuring with
<code class="literal">--enable-newlib-hw-fp</code>).
</p></td></tr></table></div><p>
For most targets it is sufficient to define just one of
<code class="literal">__IEEE_BIG_ENDIAN</code> or
<code class="literal">__IEEE_LITTLE_ENDIAN</code>. The definitions should
always be surrounded by a conditional, so they are only used when
the target architecture is selected. For example the <span class="application">OpenRISC 1000</span> is
big-endian, so we add the following to the header file.
</p><div class="informalfigure"><pre class="programlisting">
#if defined(__or32__)
#define __IEEE_BIG_ENDIAN
#endif
</pre></div></div><div class="sect2" title="4.2.2. setjmp Buffer Size"><div class="titlepage"><div><div><h3 class="title"><a id="sec_setjmp_header"></a>4.2.2.
<code class="function">setjmp</code> Buffer Size
</h3></div></div></div><p>
The implementation of <code class="function">setjmp</code> and
<code class="function">longjmp</code> made use of a buffer to hold the
machine state. The size of that buffer is architecture dependent and
specified within the <code class="filename">newlib</code> directory in
<code class="filename">libc/include/machine/setjmp.h</code>.
</p><p>
The header specifies the number of entries in the buffer and the
size of each entry (as a C type). So for the <span class="application">OpenRISC 1000</span> we use the
following.
</p><div class="informalfigure"><pre class="programlisting">
#if defined(__or32__)
/* Enough space for all regs except r0 and r11 and the status register */
#define _JBLEN 31
#define _JBTYPE unsigned long
#endif
</pre></div><p>
As before, the definition is within a conditional, so it is only
used when the target is the <span class="application">OpenRISC 1000</span> 32-bit architecture.
</p><p>
The type <span class="type">jmp_buf</span> used with
<code class="function">setjmp</code> and <code class="function">longjmp</code> is then
defined as:
</p><div class="informalfigure"><pre class="programlisting">
typedef _JBTYPE jmp_buf[_JBLEN];
</pre></div></div><div class="sect2" title="4.2.3. Miscellaneous System Definitions"><div class="titlepage"><div><div><h3 class="title"><a id="sec_newlib_config_header"></a>4.2.3.
Miscellaneous System Definitions
</h3></div></div></div><p>
Various system wide constants are specified within the
<code class="filename">newlib</code> directory in
<code class="filename">libc/include/sys/config.h</code>.
</p><p>
Very often the system default values are quite sufficient (this is
the case for the <span class="application">OpenRISC 1000</span> ). However target specific overrides of these
values can be provided at the end of this file. This file is
included in all other source files, so can be used to redefine any
of the constants used in the system. The existing file gives
numerous examples from different machines.
</p><div class="caution" title="Caution" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Caution"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Caution]" src="./images/caution.png" /></td><th align="left">Caution</th></tr><tr><td align="left" valign="top"><p>
A number of the constants defined here mirror those in <acronym class="acronym">GCC</acronym>'s
<code class="filename">limits.h</code> file. They should be kept
consistent.
</p></td></tr></table></div></div><div class="sect2" title="4.2.4. Overriding Other Header Files"><div class="titlepage"><div><div><h3 class="title"><a id="sec_other_headers"></a>4.2.4.
Overriding Other Header Files
</h3></div></div></div><p>
If other headers must be overridden (not usually necessary with a
simple port), then the new versions can be placed in
<code class="filename">libc/machine/<em class="replaceable"><code>arch</code></em>/machine</code>
within the <code class="filename">newlib</code> directory. These header files
will be used in preference to those in the standard distribution's
<code class="filename">machine</code> header directory.
</p></div></div></div><div class="chapter" title="Chapter 5. Modifying libgloss"><div class="titlepage"><div><div><h2 class="title"><a id="chap_libgloss"></a>Chapter 5.
Modifying <code class="systemitem">libgloss</code>
</h2></div></div></div><p>
Any target architecture may need multiple implementations, suited to
different platforms on which the code may run. The connection between
the library and a specific platform is known as a <a class="firstterm" href="#id2724681"><em class="firstterm">Board
Support Package</em></a> (<acronym class="acronym">BSP</acronym>). In recent versions of <code class="systemitem">newlib</code>,
<acronym class="acronym">BSP</acronym>s are separated out into their own library, <code class="systemitem">libgloss</code>, the source
for which is in the top level <code class="filename">libgloss</code> directory.
</p><p>
For <code class="systemitem">newlib</code> the <acronym class="acronym">BSP</acronym> within <code class="systemitem">libgloss</code> comprises an implementation of
the C runtime initialization, <code class="filename">crt0.o</code>, a definition
of one global data structure, and implementation of eighteen system
calls for each platform.
</p><div class="note" title="Note" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Note"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Note]" src="./images/note.png" /></td><th align="left">Note</th></tr><tr><td align="left" valign="top"><p>
<code class="systemitem">libgloss</code> is a relatively new addition to <code class="systemitem">newlib</code>. Some older ports
still have the <acronym class="acronym">BSP</acronym> code within the <code class="filename">newlib</code>
directory.
</p></td></tr></table></div><div class="sect1" title="5.1. The Platform Directory"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_platform_dir"></a>5.1.
The Platform Directory
</h2></div></div></div><p>
A directory is created in the <code class="filename">libgloss</code> directory
corresponding to the machine directory created in the
<code class="filename">newlib/libc/machine</code> directory (see <a class="xref" href="#chap_newlib" title="Chapter 4. Modifying newlib">Chapter 4</a>).
</p><p>
This directory will hold the source code for the C runtime
initialization (<code class="filename">crt0.o</code>) and for the system calls
for each <acronym class="acronym">BSP</acronym>
</p><div class="sect2" title="5.1.1. Ensuring the Platform Directory is Configured"><div class="titlepage"><div><div><h3 class="title"><a id="sec_platform_dir_config"></a>5.1.1.
Ensuring the Platform Directory is Configured
</h3></div></div></div><p>
<code class="filename">configure.in</code> within the
<code class="filename">libgloss</code> directory includes a
<code class="literal">case</code> statement configuring the target for each
target platform. This should be extended to add the new platform
directory. The <span class="application">OpenRISC 1000</span> 32-bit target requires the following change.
</p><div class="informalfigure"><pre class="programlisting">
case "${target}" in
i[[3456]]86-*-elf* | i[[3456]]86-*-coff*)
AC_CONFIG_SUBDIRS([i386])
;;
m32r-*-*)
AC_CONFIG_SUBDIRS([m32r])
;;
<Other targets not shown>
spu-*-elf)
AC_CONFIG_SUBDIRS([spu])
config_testsuite=false
config_libnosys=false
;;
or32-*-*)
AC_CONFIG_SUBDIRS(or32)
;;
iq2000-*-*)
AC_CONFIG_SUBDIRS([iq2000])
;;
esac
</pre></div><p>
After making this change the <code class="filename">configure</code> file
should be regenerated by running <span class="application">autoconf</span>.
</p></div></div><div class="sect1" title="5.2. The C Runtime Initialization, crt0.o"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_crt0"></a>5.2.
The C Runtime Initialization, <code class="filename">crt0.o</code>
</h2></div></div></div><p>
The C Runtime system must carry out the following tasks.
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
Set up the target platform in a consistent state. For example
setting up appropriate exception vectors.
</p></li><li class="listitem"><p>
Initialize the stack and frame pointers
</p></li><li class="listitem"><p>
Invoke the C constructor initialization and ensure destructors
are called on exit.
</p></li><li class="listitem"><p>
Carry out any further platform specific initialization.
</p></li><li class="listitem"><p>
Call the C <code class="function">main</code> function.
</p></li><li class="listitem"><p>
Exit with the return code supplied if the C
<code class="function">main</code> function ever terminates.
</p></li></ul></div><p>
The code is invariably assembler, although it may call out to C
functions, and is best illustrated by example from the <span class="application">OpenRISC 1000</span> . This
is a <acronym class="acronym">BSP</acronym> designed for use with a fast architectural simulator. It
comes in two variants, one providing just standard output to the
console, the other implementing a simulated <acronym class="acronym">UART</acronym> with both
standard input and standard output. The <code class="filename">crt0.0</code>
is common to both <acronym class="acronym">BSP</acronym>s and found in <code class="filename">crt0.S</code>.
</p><div class="sect2" title="5.2.1. Exception vector setup"><div class="titlepage"><div><div><h3 class="title"><a id="sec_vectors"></a>5.2.1.
Exception vector setup
</h3></div></div></div><p>
The first requirement is to populate the exception vectors. The
<span class="application">OpenRISC 1000</span> uses memory from <code class="literal">0x0</code> to
<code class="literal">0x1fff</code> for exception vectors, with vectors
placed <code class="literal">0x100</code> bytes apart. Thus a reset
exception will jump to <code class="literal">0x100</code>, a bus error
exception to <code class="literal">0x200</code> and so on.
</p><p>
In this simple <acronym class="acronym">BSP</acronym>, the vast majority of exceptions are not
supported. If they are received, they print out (using
<code class="function">printf</code>) identification of the exception and
the address which caused it to the simulator console, and then
exit. We provide a macro for that assembly code, since it will be
reused many times.
</p><div class="informalfigure"><pre class="programlisting">
#define UNHANDLED_EXCEPTION(str) \
l.addi r1,r1,-20 /* Standard prologue */ ;\
l.sw 16(r1),r2 ;\
l.addi r2,r1,20 ;\
l.sw 12(r1),r9 ;\
;\
l.movhi r3,hi(.Lfmt) /* printf format string */ ;\
l.ori r3,r3,lo(.Lfmt) ;\
l.sw 0(r1),r3 ;\
l.movhi r4,hi(str) /* Name of exception */ ;\
l.ori r4,r4,lo(str) ;\
l.sw 4(r1),r4 ;\
l.mfspr r5,r0,SPR_EPCR_BASE /* Source of the interrupt */ ;\
l.jal _printf ;\
l.sw 8(r1),r5 ;\
;\
l.ori r3,r0,0xffff /* Failure RC */ ;\
l.jal _exit ;\
l.nop ;\
;\
l.rfe /* Never executed we hope */
</pre></div><p>
The call to <code class="function">printf</code> is expected to use a
standard format string (at the label <code class="literal">.Lfmt</code>)
which requires two other arguments, an identification string
(labeled by the parameter <code class="literal">st</code> to the macro) and
the program counter where the exception occurred (loaded from
Special Purpose Register <code class="literal">SPR_EPCR_BASE</code>). Return
from exception is provided as a formality, although the call to
<code class="function">exit</code> means that we should never execute it.
</p><p>
Note that compiled C functions have their names prepended by
underscore on the <span class="application">OpenRISC 1000</span> . It is these names that must be used from
the assembler code.
</p><p>
The format and identification strings are read only data.
</p><div class="informalfigure"><pre class="programlisting">
.section .rodata
.Lfmt: .string "Unhandled %s exception at address %08p\n"
.L200: .string "bus error"
.L300: .string "data page fault"
.L400: .string "instruction page fault"
.L500: .string "timer"
.L600: .string "alignment"
.L700: .string "illegal instruction"
.L800: .string "external interrupt"
.L900: .string "data TLB"
.La00: .string "instruction TLB"
.Lb00: .string "range"
.Lc00: .string "syscall"
.Ld00: .string "floating point"
.Le00: .string "trap"
.Lf00: .string "undefined 0xf00"
.L1000: .string "undefined 0x1000"
.L1100: .string "undefined 0x1100"
.L1200: .string "undefined 0x1200"
.L1300: .string "undefined 0x1300"
.L1400: .string "undefined 0x1400"
.L1500: .string "undefined 0x1500"
.L1600: .string "undefined 0x1600"
.L1700: .string "undefined 0x1700"
.L1800: .string "undefined 0x1800"
.L1900: .string "undefined 0x1900"
.L1a00: .string "undefined 0x1a00"
.L1b00: .string "undefined 0x1b00"
.L1c00: .string "undefined 0x1c00"
.L1d00: .string "undefined 0x1d00"
.L1e00: .string "undefined 0x1e00"
.L1f00: .string "undefined 0x1f00"
</pre></div><p>
The first executable code is for the exception vectors. These must
go first in memory, so are placed in their own section,
<code class="literal">.vectors</code>. The linker/loader will ensure this
this code is placed first in memory (see <a class="xref" href="#sec_linker" title="7.3. Changes to the GNU Linker">Section 7.3</a>).
</p><p>
The reset vector just jumps to the start code. The code is too
large to sit within the <code class="literal">0x100</code> bytes of an
exception vector entry, and is placed in the main text space, in
function <code class="function">_start</code>.
</p><div class="informalfigure"><pre class="programlisting">
.section .vectors,"ax"
/* 0x100: RESET exception */
.org 0x100
_reset:
/* Jump to program initialisation code */
l.movhi r2,hi(_start)
l.ori r2,r2,lo(_start)
l.jr r2
l.nop
</pre></div><p>
The second vector, at address <code class="literal">0x200</code> is the bus
error exception vector. In normal use, like all other exceptions
it it causes exit and uses the
<code class="literal">UNHANDLED_EXCEPTION</code> macro.
</p><p>
However during start up, the code tries deliberately to write out
of memory, to determine the end of memory, which will trigger this
bus exception. For this a simple exception handler, which just
skips the offending instruction is required.
</p><p>
The solution is to place this code first, followed by the
unhandled exception code. Once the end of memory has been located,
the initial code can be overwritten by <code class="literal">l.nop</code>
opcodes, so the exception will drop through to the
<code class="literal">UNHANDLED_EXCEPTOON</code> code.
</p><div class="informalfigure"><pre class="programlisting">
.org 0x200
_buserr:
l.mfspr r24,r0,SPR_EPCR_BASE
l.addi r24,r24,4 /* Return one instruction on */
l.mtspr r0,r24,SPR_EPCR_BASE
l.rfe
_buserr_std:
UNHANDLED_EXCEPTION (.L200)
</pre></div><p>
No effort is made to save the register (<code class="literal">r24</code>) that
is used in the handler. The start up code testing for end of memory
must not use this register.
</p><p>
The next exception, data page fault, at location 0x300, like all
other exceptions is unhandled.
</p><div class="informalfigure"><pre class="programlisting">
.org 0x300
UNHANDLED_EXCEPTION (.L300)
</pre></div></div><div class="sect2" title="5.2.2. The _start Function and Stack Initialization"><div class="titlepage"><div><div><h3 class="title"><a id="sec_stack_init"></a>5.2.2.
The <code class="function">_start</code> Function and Stack Initialization
</h3></div></div></div><p>
The <span class="application">OpenRISC 1000</span> <acronym class="acronym">ABI</acronym> uses a falling stack. The linker will place code
and static data at the bottom of memory (starting with the
exception vectors). The heap then starts immediately after this,
while the stack grows down from the end of memory.
</p><p>
The linker will supply the address for the start of heap (it is in
the global variable <code class="varname">end</code>). However we must find
the stack location by trying to write to memory above the heap to
determine the end of memory. Rather than write to every location,
the code assumes memory is a multiple of 64KB, and tries writing
to the last word of each 64KB block above <code class="literal">end</code>
until the value read back fails.
</p><p>
This failure will trigger a bus error exception, which must be
handled (see <a class="xref" href="#sec_vectors" title="5.2.1. Exception vector setup">Section 5.2.1</a>). The address used for
the start of the stack (which is also the last word of memory) is
stored in a global location, <code class="varname">_stack</code> (which C
will recognize as <code class="varname">stack</code>).
</p><div class="informalfigure"><pre class="programlisting">
.section .data
.global _stack
_stack: .space 4,0
</pre></div><p>
<code class="function">_start</code> is declared so it looks like a C
function. <acronym class="acronym">GDB</acronym> knows that <code class="function">_start</code> is special,
and this will ensure that backtraces do not wind back further than
<code class="function">main</code>. It is located in ordinary text space,
so will be placed with other code by the linker/loader.
</p><div class="informalfigure"><pre class="programlisting">
.section .text
.global _start
.type _start,@function
_start:
</pre></div><p>
The first memory location to test is found by rounding the
<code class="varname">end</code> location down to a multiple of 64KB, then
taking the last word of the 64KB above
that. <code class="literal">0xaaaaaaaa</code> is used as the test word to
write to memory and read back.
</p><div class="informalfigure"><pre class="programlisting">
l.movhi r30,hi(end)
l.ori r30,r30,lo(end)
l.srli r30,r30,16 /* Round down to 64KB boundary */
l.slli r30,r30,16
l.addi r28,r0,1 /* Constant 64KB in register */
l.slli r28,r28,16
l.add r30,r30,r28
l.addi r30,r30,-4 /* SP one word inside next 64KB? */
l.movhi r26,0xaaaa /* Test pattern to store in memory */
l.ori r26,r26,0xaaaa
</pre></div><p>
Each 64KB block is tested by writing the test value and reading
back to see if it matches.
</p><div class="informalfigure"><pre class="programlisting">
.L3:
l.sw 0(r30),r26
l.lwz r24,0(r30)
l.sfeq r24,r26
l.bnf .L4
l.nop
l.j .L3
l.add r30,r30,r28 /* Try 64KB higher */
.L4:
</pre></div><p>
The previous value is then the location to use for end of stack,
and should be stored in the <code class="varname">_stack</code> location.
</p><div class="informalfigure"><pre class="programlisting">
l.sub r30,r30,r28 /* Previous value was wanted */
l.movhi r26,hi(_stack)
l.ori r26,r26,lo(_stack)
l.sw 0(r26),r30
</pre></div><p>
The stack pointer (<code class="literal">r1</code>) and frame pointer
(<code class="literal">r2</code>) can be initialized with this value.
</p><div class="informalfigure"><pre class="programlisting">
l.add r1,r30,r0
l.add r2,r30,r0
</pre></div><p>
Having determined the end of memory, there is no need to handle
bus errors silently. The words of code between
<code class="literal">_buserr</code> and <code class="literal">_buserr_std</code> can
be replaced by <code class="literal">l.nop</code>.
</p><div class="informalfigure"><pre class="programlisting">
l.movhi r30,hi(_buserr)
l.ori r30,r30,lo(_buserr)
l.movhi r28,hi(_buserr_std)
l.ori r28,r28,lo(_buserr_std)
l.movhi r26,0x1500 /* l.nop 0 */
l.ori r26,r26,0x0000
.L5:
l.sfeq r28,r30
l.bf .L6
l.nop
l.sw 0(r30),r26 /* Patch the instruction */
l.j .L5
l.addi r30,r30,4 /* Next instruction */
.L6:
</pre></div><div class="note" title="Note" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Note"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Note]" src="./images/note.png" /></td><th align="left">Note</th></tr><tr><td align="left" valign="top"><p>
It is essential that this code is before any data or instruction
cache is initialized. Otherwise more complex steps would be
required to enforce data write back and invalidate any
instruction cache entry.
</p></td></tr></table></div></div><div class="sect2" title="5.2.3. Cache Initialization"><div class="titlepage"><div><div><h3 class="title"><a id="id2717901"></a>5.2.3.
Cache Initialization
</h3></div></div></div><p>
The OpenRISC 1000 has optional instruction and data caches. If
these are declared (in the <code class="filename">or1ksim-board.h</code>
header), then they must be enabled by setting the appropriate bit
in the <a class="firstterm" href="#id2724791"><em class="firstterm">supervision register</em></a>.
</p><p>
This is an example of machine specific initialization.
</p><div class="informalfigure"><pre class="programlisting">
/* Cache initialisation. Enable IC and/or DC */
.if IC_ENABLE || DC_ENABLE
l.mfspr r10,r0,SPR_SR
.if IC_ENABLE
l.ori r10,r10,SPR_SR_ICE
.endif
.if DC_ENABLE
l.ori r10,r10,SPR_SR_DCE
.endif
l.mtspr r0,r10,SPR_SR
l.nop /* Flush the pipeline. */
l.nop
l.nop
l.nop
l.nop
.endif
</pre></div></div><div class="sect2" title="5.2.4. Clearing BSS"><div class="titlepage"><div><div><h3 class="title"><a id="id2717944"></a>5.2.4.
Clearing <acronym class="acronym">BSS</acronym>
</h3></div></div></div><p>
<a class="firstterm" href="#gloss_bss"><em class="firstterm"><acronym class="acronym">BSS</acronym></em></a> is the area of
memory used to hold static variables which must be initialized to
zero. Its start and end are defined by two variables from the
linker/loader, <code class="varname">__bss_start</code> and
<code class="varname">end</code> respectively.
</p><div class="informalfigure"><pre class="programlisting">
l.movhi r28,hi(__bss_start)
l.ori r28,r28,lo(__bss_start)
l.movhi r30,hi(end)
l.ori r30,r30,lo(end)
.L1:
l.sw (0)(r28),r0
l.sfltu r28,r30
l.bf .L1
l.addi r28,r28,4
</pre></div></div><div class="sect2" title="5.2.5. Constructor and Destructor Handling"><div class="titlepage"><div><div><h3 class="title"><a id="id2717990"></a>5.2.5.
Constructor and Destructor Handling
</h3></div></div></div><p>
<acronym class="acronym">GCC</acronym> may require constructors to be initialized at start up and
destructors to be called on exit. This behavior is captured in the
<acronym class="acronym">GCC</acronym> functions <code class="function">__do_global_ctors</code> and
<code class="function">__do_global_dtors</code>. There is some complexity
associated with this functionality, since there may be separate
lists for the main code and shared libraries that are dynamically
loaded.
</p><p>
It is usual to wrap this functionality in two functions,
<code class="function">init</code> and <code class="function">fini</code>, which are
placed in their own sections, <code class="literal">.init</code> and
<code class="literal">.fini</code>. The <code class="literal">.init</code> section is
loaded before all other text sections and the
<code class="literal">.fini</code> section after all other text sections.
</p><p>
The start up code should call <code class="function">init</code> to handle
any constructors.
</p><div class="informalfigure"><pre class="programlisting">
l.jal init
l.nop
</pre></div><p>
The <code class="function">fini</code> function is passed to the library
function <code class="function">_atexit</code> to ensure it is called on a
normal exit.
</p><div class="informalfigure"><pre class="programlisting">
l.movhi r3,hi(fini)
l.jal _atexit
l.ori r3,r3,lo(fini) /* Delay slot */
</pre></div></div><div class="sect2" title="5.2.6. C Initialization Functions"><div class="titlepage"><div><div><h3 class="title"><a id="id2718120"></a>5.2.6.
C Initialization Functions
</h3></div></div></div><p>
Now that the C infrastructure is set up, it is appropriate to call
any C functions that are used during initialization. In the <span class="application">OpenRISC 1000</span>
case this is a function to initialize a <acronym class="acronym">UART</acronym>. Only one version
of the library actually has a <acronym class="acronym">UART</acronym>. However it is easiest to
substitute a dummy version of the initialization function in the
version of the library without a <acronym class="acronym">UART</acronym>, rather than making this
function conditional.
</p><div class="informalfigure"><pre class="programlisting">
l.jal __uart_init
l.nop
</pre></div></div><div class="sect2" title="5.2.7. Invoking the main program"><div class="titlepage"><div><div><h3 class="title"><a id="id2718170"></a>5.2.7.
Invoking the main program
</h3></div></div></div><p>
The final stage is to call the main program. In this simple
implementation there is no mechanism to pass arguments or
environments to <code class="function">main</code>, so the arguments
<code class="varname">argc</code>, <code class="varname">argv</code> and
<code class="varname">env</code> (in <code class="literal">r3</code>,
<code class="literal">r4</code> and <code class="literal">r5</code>) are set to
<code class="literal">0</code>, <code class="literal">NULL</code> and
<code class="literal">NULL</code> respectively.
</p><div class="informalfigure"><pre class="programlisting">
l.or r3,r0,r0
l.or r4,r0,r0
l.jal _main
l.or r5,r0,r0 /* Delay slot */
</pre></div><p>
If the main program returns, its result (held in
<code class="literal">r11</code> on the <span class="application">OpenRISC 1000</span> ) will be a return code from
the program, which we pass to the <code class="function">exit</code>.
</p><div class="informalfigure"><pre class="programlisting">
l.jal _exit
l.addi r3,r11,0 /* Delay slot */
</pre></div><p>
<code class="function">exit</code> should not return, but just in case, we
can put the processor in a tight loop at this stage, in order to
ensure consistent behavior.
</p><div class="informalfigure"><pre class="programlisting">
.L2:
l.j .L2
l.nop
</pre></div></div></div><div class="sect1" title="5.3. Standard System Call Implementations"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_syscalls"></a>5.3.
Standard System Call Implementations
</h2></div></div></div><p>
The simplest way to provide a board support package is to implement
the 18 system calls in non-reentrant fashion. For many bare metal
implementations this is sufficient.
</p><p>
The simplest possible <acronym class="acronym">BSP</acronym> supports just output to standard output
and non input. We give the minimal implementation for such a system.
</p><p>
Where appropriate, we also show the <span class="application">OpenRISC 1000</span> implementation as a
practical example.
</p><p>
This section duplicates much of the information found in the
<code class="systemitem">newlib</code> <code class="systemitem">libc</code> documentation <a class="xref" href="#ref_libc" title="The Red Hat Newlib C Library">[1]</a>. It is
included here for completeness.
</p><div class="sect2" title="5.3.1. Error Handling"><div class="titlepage"><div><div><h3 class="title"><a id="id2718360"></a>5.3.1.
Error Handling
</h3></div></div></div><p>
Many functions set an error code on failure in the global
variable, <code class="varname">errno</code>.
</p><p>
There is a slight complication with <code class="systemitem">newlib</code>, because
<code class="varname">errno</code> is not implemented as a variable, but a
macro (this make life easier for reentrant functions).
</p><p>
The solution for standard system call implementations, which must
return an error code is to undefine the macro and use the external
variable instead. At the head of such functions use the following.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
</pre></div><div class="note" title="Note" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Note"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Note]" src="./images/note.png" /></td><th align="left">Note</th></tr><tr><td align="left" valign="top"><p>
<code class="varname">errno</code> is a global variable, so changing it
will immediately make a function non-reentrant.
</p></td></tr></table></div></div><div class="sect2" title="5.3.2. The Global Environment, environ"><div class="titlepage"><div><div><h3 class="title"><a id="id2718422"></a>5.3.2.
The Global Environment, <code class="varname">environ</code>
</h3></div></div></div><p>
The global variable, <code class="varname">environ</code> must point to a
null terminated list of environment variable name-value pairs.
</p><p>
For a minimal implementation it is sufficient to use an empty list
as follows.
</p><div class="informalfigure"><pre class="programlisting">
char *__env[1] = { 0 };
char **environ = __env;
</pre></div></div><div class="sect2" title="5.3.3. Exit a program, _exit"><div class="titlepage"><div><div><h3 class="title"><a id="sec_exit"></a>5.3.3.
Exit a program, <code class="function">_exit</code>
</h3></div></div></div><p>
Exit a program without any cleanup.
</p><p>
The <span class="application">OpenRISC 1000</span> s implementation makes use of the
<code class="literal">l.nop</code> opcode. This opcode takes a 16-bit
immediate operand. Functionally the operand has no effect on the
processor itself. However a simulator can inspect the operand to
provide additional behavior external to the machine.
</p><p>
When executing on <span class="application">Or1ksim</span>, <code class="literal">l.nop 1</code> causes a
tidy exit of the simulator, using the value in
<code class="literal">r3</code> as the return code.
</p><div class="informalfigure"><pre class="programlisting">
void
_exit (int rc)
{
register int t1 asm ("r3") = rc;
asm volatile ("\tl.nop\t%0" : : "K" (NOP_EXIT), "r" (t1));
while (1)
{
}
} /* _exit () */
</pre></div><p>
Note the use of <code class="literal">volatile</code>. Otherwise there is a
strong possibility of an optimizing compiler recognizing that this
opcode does nothing (we are relying on a simulation side-effect)
and removing it.
</p><div class="caution" title="Caution" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Caution"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Caution]" src="./images/caution.png" /></td><th align="left">Caution</th></tr><tr><td align="left" valign="top"><p>
The name of this function is already namespace clean. If a
namespace clean implementation of the system calls has been
specified in <code class="filename">configure.host</code> (see <a class="xref" href="#sec_configure_host" title="3.3.1. Extending configure.host for a New Target">Section 3.3.1</a>), then this function is still
named <code class="function">_exit</code>, not
<code class="function">__exit</code>.
</p></td></tr></table></div></div><div class="sect2" title="5.3.4. Closing a file, close"><div class="titlepage"><div><div><h3 class="title"><a id="id2718572"></a>5.3.4.
Closing a file, <code class="function">close</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_close</code>, otherwise implement
<code class="function">close</code>. The detailed implementation will
depend on the file handling functionality available.
</p><p>
In the minimal implementation, this function always fails, since
there is only standard output, which is not a valid file to
close. This implementation is sufficient for the <span class="application">OpenRISC 1000</span> .
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
int
_close (int file)
{
errno = EBADF;
return -1; /* Always fails */
} /* _close () */
</pre></div></div><div class="sect2" title="5.3.5. Transfer Control to a New Process, execve"><div class="titlepage"><div><div><h3 class="title"><a id="id2718631"></a>5.3.5.
Transfer Control to a New Process, <code class="function">execve</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_execve</code>, otherwise implement
<code class="function">execve</code>. The implementation of this
functionality will be tightly bound to any operating
infrastructure for handling multiple processes.
</p><p>
A minimal implementation, such as that for bare metal coding, only
offers a single user thread of control. It is thus impossible to
start a new process, so this function always fails.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno;
extern int errno;
int
_execve (char *name,
char **argv,
char **env)
{
errno = ENOMEM;
return -1; /* Always fails */
} /* _execve () */
</pre></div><p>
The choice of <code class="literal">errno</code> is somewhat
arbitrary. However no value for "no processes available" is
provided, and <code class="literal">ENOMEM</code> is the closest in meaning
to this.
</p></div><div class="sect2" title="5.3.6. Create a new process, fork"><div class="titlepage"><div><div><h3 class="title"><a id="id2718701"></a>5.3.6.
Create a new process, <code class="function">fork</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_fork</code>, otherwise implement
<code class="function">fork</code>. The implementation of this
functionality will be tightly bound to any operating
infrastructure for handling multiple processes.
</p><p>
A minimal implementation, such as that for bare metal coding, only
offers a single user thread of control. It is thus impossible to
start a new process, so this function always fails.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
int
_fork ()
{
errno = EAGAIN;
return -1; /* Always fails */
} /* _fork () */
</pre></div><p>
The choice of <code class="literal">errno</code> is again somewhat
arbitrary. However no value for "no processes available" is
provided, and <code class="literal">EAGAIN</code> is the closest in meaning
to this.
</p></div><div class="sect2" title="5.3.7. Provide the Status of an Open File, fstat"><div class="titlepage"><div><div><h3 class="title"><a id="id2718768"></a>5.3.7.
Provide the Status of an Open File, <code class="function">fstat</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_fstat</code>, otherwise implement
<code class="function">fstat</code>. The detailed implementation will
depend on the file handling functionality available.
</p><p>
A minimal implementation should assume that all files are
character special devices and populate the status data structure
accordingly.
</p><div class="informalfigure"><pre class="programlisting">
#include <sys/stat.h>
int
_fstat (int file,
struct stat *st)
{
st->st_mode = S_IFCHR;
return 0;
} /* _fstat () */
</pre></div><p>
The <span class="application">OpenRISC 1000</span> implementation requires two versions of this, one for
the <acronym class="acronym">BSP</acronym> using the console for output and one for the <acronym class="acronym">BSP</acronym> using
a <acronym class="acronym">UART</acronym> and supporting both standard input and standard output.
</p><p>
Without a <acronym class="acronym">UART</acronym>, the implementation still checks that the file
descriptor is one of the two that are supported, and otherwise
returns an error.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#include <sys/stat.h>
#include <unistd.h>
#undef errno
extern int errno;
int
_fstat (int file,
struct stat *st)
{
if ((STDOUT_FILENO == file) || (STDERR_FILENO == file))
{
st->st_mode = S_IFCHR;
return 0;
}
else
{
errno = EBADF;
return -1;
}
} /* _fstat () */
</pre></div><p>
The implementation when a <acronym class="acronym">UART</acronym> is available is almost identical,
except that <code class="literal">STDIN_FILENO</code> is also an acceptable
file for which status can be provided.
</p></div><div class="sect2" title="5.3.8. Get the Current Process ID, getpid"><div class="titlepage"><div><div><h3 class="title"><a id="id2718888"></a>5.3.8.
Get the Current Process ID, <code class="function">getpid</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_getpid</code>, otherwise implement
<code class="function">getpid</code>. The implementation of this
functionality will be tightly bound to any operating
infrastructure for handling multiple processes.
</p><p>
For a minimal implementation, with no processes, this can just
return a constant. It is perhaps safer to return one rather than
zero, to avoid issue with software that believes process zero is
something special.
</p><div class="informalfigure"><pre class="programlisting">
int
_getpid ()
{
return 1; /* Success */
} /* _getpid () */
</pre></div></div><div class="sect2" title="5.3.9. Determine the Nature of a Stream, isatty"><div class="titlepage"><div><div><h3 class="title"><a id="id2718937"></a>5.3.9.
Determine the Nature of a Stream, <code class="function">isatty</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_isatty</code>, otherwise implement
<code class="function">isatty</code>. The detailed implementation will
depend on the file handling functionality available.
</p><p>
This specifically checks whether a stream is a terminal. The
minimal implementation only has the single output stream, which is
to the console, so always returns 1.
</p><div class="informalfigure"><pre class="programlisting">
int
_isatty (int file)
{
return 1;
} /* _isatty () */
</pre></div><div class="caution" title="Caution" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Caution"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Caution]" src="./images/caution.png" /></td><th align="left">Caution</th></tr><tr><td align="left" valign="top"><p>
Contrary to the standard <code class="systemitem">libc</code> documentation, this applies to
any stream, not just output streams.
</p></td></tr></table></div><p>
The <span class="application">OpenRISC 1000</span> version gives a little more detail, setting
<code class="varname">errno</code> if the stream is not standard output,
standard error or (for the <acronym class="acronym">UART</acronym> version of the <acronym class="acronym">BSP</acronym>) standard
input.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#include <unistd.h>
#undef ERRNO
extern int errno;
int
_isatty (int file)
{
if ((file == STDOUT_FILENO) || (file == STDERR_FILENO))
{
return 1;
}
else
{
errno = EBADF;
return -1;
}
} /* _isatty () */
</pre></div><p>
The <acronym class="acronym">UART</acronym> version is almost identical, but also succeeds for
standard input.
</p></div><div class="sect2" title="5.3.10. Send a Signal, kill"><div class="titlepage"><div><div><h3 class="title"><a id="id2719051"></a>5.3.10.
Send a Signal, <code class="function">kill</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_kill</code>, otherwise implement
<code class="function">kill</code>. The implementation of this
functionality will be tightly bound to any operating
infrastructure for handling multiple processes.
</p><p>
A minimal implementation has no concept of either signals, nor of
processes to receive those signals. So this function should always
fail with an appropriate value in <code class="varname">errno</code>.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
int
_kill (int pid,
int sig)
{
errno = EINVAL;
return -1; /* Always fails */
} /* _kill () */
</pre></div></div><div class="sect2" title="5.3.11. Rename an existing file, link"><div class="titlepage"><div><div><h3 class="title"><a id="id2719106"></a>5.3.11.
Rename an existing file, <code class="function">link</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_link</code>, otherwise implement
<code class="function">link</code>. The detailed implementation will
depend on the file handling functionality available.
</p><p>
A minimal implementation has no file system, so this function must
always fail, with an appropriate value set in
<code class="varname">errno</code>.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
int
_link (char *old,
char *new)
{
errno = EMLINK;
return -1; /* Always fails */
} /* _link () */
</pre></div></div><div class="sect2" title="5.3.12. Set Position in a File, lseek"><div class="titlepage"><div><div><h3 class="title"><a id="id2719160"></a>5.3.12.
Set Position in a File, <code class="function">lseek</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_lseek</code>, otherwise implement
<code class="function">lseek</code>. The detailed implementation will
depend on the file handling functionality available.
</p><p>
A minimal implementation has no file system, so this function can
return 0, indicating that the only stream (standard output) is
positioned at the start of file.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
int
_lseek (int file,
int offset,
int whence)
{
return 0;
} /* _lseek () */
</pre></div><p>
The <span class="application">OpenRISC 1000</span> version is a little more detailed, returning zero only
if the stream is standard output, standard error or (for the
<acronym class="acronym">UART</acronym> version of the <acronym class="acronym">BSP</acronym>) standard input. Otherwise -1 is
returned and an appropriate error code set in
<code class="varname">errno</code>.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#include <unistd.h>
#undef errno
extern int errno;
int
_lseek (int file,
int offset,
int whence)
{
if ((STDOUT_FILENO == file) || (STDERR_FILENO == file))
{
return 0;
}
else
{
errno = EBADF;
return (long) -1;
}
} /* _lseek () */
</pre></div><p>
The <acronym class="acronym">UART</acronym> version is almost identical, but also succeeds for
standard input.
</p></div><div class="sect2" title="5.3.13. Open a file, open"><div class="titlepage"><div><div><h3 class="title"><a id="id2719266"></a>5.3.13.
Open a file, <code class="function">open</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_open</code>, otherwise implement
<code class="function">open</code>. The detailed implementation will
depend on the file handling functionality available.
</p><p>
A minimal implementation has no file system, so this function must
always fail, with an appropriate error code set in
<code class="varname">errno</code>.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
int
_open (const char *name,
int flags,
int mode)
{
errno = ENOSYS;
return -1; /* Always fails */
} /* _open () */
</pre></div></div><div class="sect2" title="5.3.14. Read from a File, read"><div class="titlepage"><div><div><h3 class="title"><a id="id2719321"></a>5.3.14.
Read from a File, <code class="function">read</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_read</code>, otherwise implement
<code class="function">read</code>. The detailed implementation will depend
on the file handling functionality available.
</p><p>
A minimal implementation has no file system. Rather than failing,
this function returns 0, indicating end-of-file.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
int
_read (int file,
char *ptr,
int len)
{
return 0; /* EOF */
} /* _read () */
</pre></div><p>
The <span class="application">OpenRISC 1000</span> <acronym class="acronym">BSP</acronym> without a <acronym class="acronym">UART</acronym> is very similar to the minimal
implementation, but checks that the stream is standard input
before returning 0. For all other streams it returns an error.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#include <unistd.h>
#undef errno
extern int errno;
int
_read (int file,
char *ptr,
int len)
{
if (STDIN_FILENO == file)
{
return 0; /* EOF */
}
else
{
errno = EBADF;
return -1;
}
} /* _read () */
</pre></div><p>
The <span class="application">OpenRISC 1000</span> <acronym class="acronym">BSP</acronym> with a <acronym class="acronym">UART</acronym> is more complex. In this case, if
the stream is standard input, a character is read (and optionally
echoed) from the <acronym class="acronym">UART</acronym>.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#include <unistd.h>
#undef errno
extern int errno;
int
_read (int file,
char *buf,
int len)
{
if (STDIN_FILENO == file)
{
int i;
for (i = 0; i < len; i++)
{
buf[i] = _uart_getc ();
#ifdef UART_AUTO_ECHO
_uart_putc (buf[i]);
#endif
/* Return partial buffer if we get EOL */
if ('\n' == buf[i])
{
return i;
}
}
return i; /* Filled the buffer */
}
else
{
errno = EBADF;
return -1;
}
} /* _read () */
</pre></div><div class="caution" title="Caution" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Caution"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Caution]" src="./images/caution.png" /></td><th align="left">Caution</th></tr><tr><td align="left" valign="top"><p>
The <span class="application">Or1ksim</span> <acronym class="acronym">UART</acronym> implementation only returns data when
carriage return is hit, rather than as each character becomes
available, which can lead to some unexpected behavior.
</p></td></tr></table></div></div><div class="sect2" title="5.3.15. Allocate more Heap, sbrk"><div class="titlepage"><div><div><h3 class="title"><a id="sec_sbrk"></a>5.3.15.
Allocate more Heap, <code class="function">sbrk</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_sbrk</code>, otherwise implement
<code class="function">sbrk</code>. This is one function for which there is
no default minimal implementation. It is important that it is
implemented wherever possible, since <code class="function">malloc</code>
depends on it, and in turn many other functions depend on
<code class="function">malloc</code>. In this application note, the <span class="application">OpenRISC 1000</span>
implementation is used as an example.
</p><p>
As noted earlier (<a class="xref" href="#sec_stack_init" title="5.2.2. The _start Function and Stack Initialization">Section 5.2.2</a>), the heap on
the <span class="application">OpenRISC 1000</span> grows up from the end of loaded program space, and the
stack grows down from the top of memory. The linker defines the
symbol <code class="varname">_end</code>, which will be the start of the
heap, whilst the C runtime initialization places the address of
the last work in memory in the global variable
<code class="varname">_stack</code>.
</p><div class="caution" title="Caution" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Caution"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Caution]" src="./images/caution.png" /></td><th align="left">Caution</th></tr><tr><td align="left" valign="top"><p>
<code class="literal">_end</code> is a symbol defined by the linker, not a
variable, so it is its <span class="emphasis"><em>address</em></span> that must be
used, not its value.
</p></td></tr></table></div><p>
Within a C program these two variables are referred to without
their leading underscore—the C compiler prepends all
variable names with underscore.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
#define STACK_BUFFER 65536 /* Reserved stack space in bytes. */
void *
_sbrk (int nbytes)
{
/* Symbol defined by linker map */
extern int end; /* start of free memory (as symbol) */
/* Value set by crt0.S */
extern void *stack; /* end of free memory */
/* The statically held previous end of the heap, with its initialization. */
static void *heap_ptr = (void *)&end; /* Previous end */
if ((stack - (heap_ptr + nbytes)) > STACK_BUFFER )
{
void *base = heap_ptr;
heap_ptr += nbytes;
return base;
}
else
{
errno = ENOMEM;
return (void *) -1;
}
} /* _sbrk () */
</pre></div><p>
The program always tries to keep a minimum of 65,536
(2<sup>16</sup>) bytes spare for the stack.
</p><div class="note" title="Note" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Note"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Note]" src="./images/note.png" /></td><th align="left">Note</th></tr><tr><td align="left" valign="top"><p>
This implementation defines <code class="function">_sbrk</code> as
returning type <span class="type">void *</span>. The standard newlib
documentation uses return type <span class="type">caddr_t</span>, which is
defined in <code class="filename">unistd.h</code>. The author believes
that <span class="type">void *</span> is now the recommended return type for
this function.
</p></td></tr></table></div><div class="important" title="Important" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Important"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Important]" src="./images/important.png" /></td><th align="left">Important</th></tr><tr><td align="left" valign="top"><p>
<code class="function">sbrk</code> has to return the previous end of the
heap, whose value is held in the static variable,
<code class="varname">heap_ptr</code>.
</p><p>
The problem is that this now makes the function
non-reentrant. If the function were interrupted after the
assignment to <code class="varname">base</code>, but before the following
assignment to <code class="varname">heap_ptr</code>, and the interrupt
routine itself also called <code class="function">sbrk</code>, then the
heap would become corrupted.
</p><p>
For simple systems, it would be sufficient to avoid using this
function in interrupt service routines. However the problem then
knowing which functions might call <code class="function">malloc</code>
and hence <code class="function">sbrk</code>, so effectively all library
functions must be avoided.
</p><p>
The problem cannot even be completely avoided by using reentrant
functions (see <a class="xref" href="#sec_reentrant_syscalls" title="5.4. Reentrant System Call Implementations">Section 5.4</a>), since
just providing a per thread data structure does not help. The
end of heap is a single global value. The only full solution is
to surround the update of the global variable by a semaphore,
and failing the allocation if the region is blocked (we cannot
wait, or deadlock would result).
</p></td></tr></table></div></div><div class="sect2" title="5.3.16. Status of a File (by Name), stat"><div class="titlepage"><div><div><h3 class="title"><a id="id2719729"></a>5.3.16.
Status of a File (by Name), <code class="function">stat</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_stat</code>, otherwise implement
<code class="function">stat</code>. The detailed implementation will
depend on the file handling functionality available.
</p><p>
A minimal implementation should assume that all files are
character special devices and populate the status data structure
accordingly.
</p><div class="informalfigure"><pre class="programlisting">
#include <sys/stat.h>
int
_stat (char *file,
struct stat *st)
{
st->st_mode = S_IFCHR;
return 0;
} /* _stat () */
</pre></div><p>
The <span class="application">OpenRISC 1000</span> implementation takes a stricter view of this. Since no
named files are supported, this function always fails.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#include <sys/stat.h>
#undef errno
extern int errno;
int
_stat (char *file,
struct stat *st)
{
errno = EACCES;
return -1;
} /* _stat () */
</pre></div></div><div class="sect2" title="5.3.17. Provide Process Timing Information, times"><div class="titlepage"><div><div><h3 class="title"><a id="id2719810"></a>5.3.17.
Provide Process Timing Information, <code class="function">times</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_times</code>, otherwise implement
<code class="function">times</code>. The implementation of this
functionality will be tightly bound to any operating
infrastructure for handling multiple processes.
</p><p>
A minimal implementation need not offer any timing information, so
should always fail with an appropriate value in
<code class="varname">errno</code>.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#include <sys/times.h>
#undef errno
extern int errno;
int
_times (struct tms *buf)
{
errno = EACCES;
return -1;
} /* _times () */
</pre></div></div><div class="sect2" title="5.3.18. Remove a File's Directory Entry, unlink"><div class="titlepage"><div><div><h3 class="title"><a id="id2719864"></a>5.3.18.
Remove a File's Directory Entry, <code class="function">unlink</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_unlink</code>, otherwise implement
<code class="function">unlink</code>. The detailed implementation will
depend on the file handling functionality available.
</p><p>
A minimal implementation has no file system, so this function
should always fail, setting an appropriate value in
<code class="varname">errno</code>.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
int
_unlink (char *name)
{
errno = ENOENT;
return -1; /* Always fails */
} /* _unlink () */
</pre></div></div><div class="sect2" title="5.3.19. Wait for a Child Process, wait"><div class="titlepage"><div><div><h3 class="title"><a id="id2719918"></a>5.3.19.
Wait for a Child Process, <code class="function">wait</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_wait</code>, otherwise implement
<code class="function">wait</code>. The implementation of this
functionality will be tightly bound to any operating
infrastructure for handling multiple processes.
</p><p>
A minimal implementation has only one process, so can wait for no
other process and should always fail with an appropriate value in
<code class="varname">errno</code>.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#undef errno
extern int errno;
int
_wait (int *status)
{
errno = ECHILD;
return -1; /* Always fails */
} /* _wait () */
</pre></div></div><div class="sect2" title="5.3.20. Write to a File, write"><div class="titlepage"><div><div><h3 class="title"><a id="id2719973"></a>5.3.20.
Write to a File, <code class="function">write</code>
</h3></div></div></div><p>
For a namespace clean function, implement
<code class="function">_write</code>, otherwise implement
<code class="function">write</code>. The detailed implementation will
depend on the file handling functionality available.
</p><p>
A minimal implementation only supports writing to standard
output. The core of the implementation is:
</p><div class="informalfigure"><pre class="programlisting">
int
_write (int file,
char *buf,
int nbytes)
{
int i;
/* Output character at at time */
for (i = 0; i < nbytes; i++)
{
outbyte (buf[i]);
}
return nbytes;
} /* _write () */
</pre></div><p>
The function <code class="function">outbyte</code> must use the
functionality of the target platform to write a single character
to standard output. For example copying the character to a serial
line for display. There can be no standard implementation of this
function.
</p><p>
For the <span class="application">OpenRISC 1000</span> two versions are needed one for the <acronym class="acronym">BSP</acronym> without a
<acronym class="acronym">UART</acronym> one for the <acronym class="acronym">BSP</acronym> with a <acronym class="acronym">UART</acronym>.
</p><p>
Without a <acronym class="acronym">UART</acronym> the implementation uses the
<code class="literal">l.nop</code> opcode with a parameter, as with the
implementation of <code class="function">_exit</code> (<a class="xref" href="#sec_exit" title="5.3.3. Exit a program, _exit">Section 5.3.3</a>). In this case the parameter 4 will cause the
simulator to print out the value in register <code class="literal">r3</code>
as an <acronym class="acronym">ASCII</acronym> character.
</p><div class="informalfigure"><pre class="programlisting">
#include "or1ksim-board.h"
static void
outbyte (char c)
{
register char t1 asm ("r3") = c;
asm volatile ("\tl.nop\t%0" : : "K" (NOP_PUTC), "r" (t1));
} /* outbyte () */
</pre></div><p>
We also use a stricter implementation of the main
<code class="function">write</code> function, only permitting a write if
the standard output or standard error stream is specified.
</p><div class="informalfigure"><pre class="programlisting">
#include <errno.h>
#include <unistd.h>
#undef errno
extern int errno;
int
_write (int file,
char *buf,
int nbytes)
{
int i;
/* We only handle stdout and stderr */
if ((file != STDOUT_FILENO) && (file != STDERR_FILENO))
{
errno = EBADF;
return -1;
}
/* Output character at at time */
for (i = 0; i < nbytes; i++)
{
outbyte (buf[i]);
}
return nbytes;
} /* _write () */
</pre></div><p>
For the <acronym class="acronym">BSP</acronym> supporting a <acronym class="acronym">UART</acronym>, all that is needed is to change
the <code class="function">outbyte</code> function to use the routines to
drive the <acronym class="acronym">UART</acronym>
</p><div class="informalfigure"><pre class="programlisting">
static void
outbyte (char c)
{
_uart_putc (c);
} /* outbyte () */
</pre></div><p>
The <acronym class="acronym">UART</acronym> support routines are provided separately, driving the
interface via its memory mapped registers.
</p></div></div><div class="sect1" title="5.4. Reentrant System Call Implementations"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_reentrant_syscalls"></a>5.4.
Reentrant System Call Implementations
</h2></div></div></div><p>
Reentrancy is achieved by providing a global reentrancy structure,
<span class="type">struct _reent</span> for each thread of control, which
holds thread specific versions of global data structures, such as
<code class="varname">errno</code>.
</p><p>
For a fully reentrant system, the <acronym class="acronym">BSP</acronym> should implement the
reentrant versions of the system calls, having defined
<code class="literal">syscall_dir=syscalls</code> and added
<code class="literal">-DREENTRANT_SYSCALLS_PROVIDED"</code> to
<code class="varname">newlib_cflags</code> in
<code class="filename">configure.host</code> (see <a class="xref" href="#sec_configure_host" title="3.3.1. Extending configure.host for a New Target">Section 3.3.1</a>).
</p><p>
16 of the system calls have reentrant versions, which take the
suffix <code class="literal">_r</code> and are passed an additional first
argument, which is a pointer to the reentrancy structure,
<span class="type">struct reent</span> for the thread of control. Thus
<code class="function">_close</code> is replaced by
<code class="function">_close_r</code>. The reentrant functions are
<code class="function">_close_r</code>, <code class="function">_execve_r</code>,
<code class="function">_fcntl_r</code>, <code class="function">_fork_r</code>,
<code class="function">_fstat_r</code>, <code class="function">_getpid_r</code>,
<code class="function">_link_r</code>, <code class="function">_lseek_r</code>,
<code class="function">_open_r</code>, <code class="function">_read_r</code>,
<code class="function">_sbrk_r</code>, <code class="function">_stat_r</code>,
<code class="function">_times_r</code>, <code class="function">_unlink_r</code>,
<code class="function">_wait_r</code> and <code class="function">_write_r</code>.
</p><p>
Two system calls do not need reentrant versions,
<code class="function">_kill</code> and <code class="function">_exit</code>, which are
provided as with non-reentrant versions.
</p><p>
For many of the reentrant functions, the behavior is almost
identical to that of the non-reentrant versions, beyond ensuring the
thread specific version of <code class="varname">errno</code> in the
reentrancy structure is used. Template versions can be found in the
<code class="filename">libc/reent</code> directory under the
<code class="filename">newlib</code> directory.
</p><p>
There are two ways in which the end user can be supported with these
reentrancy functions. In the first it is up to the user to manage
per thread reentrancy data structures and to call the reentrant
functions explicitly.
</p><p>
However the more powerful solution is for the system to manage the
reentrancy structure itself. The end user can call the standard
functions, and they will be mapped to reentrant calls, passing in a
reentrancy structure for the thread.
</p><p>
For this approach to be used, <code class="literal">-D__DYNAMIC_REENT__</code>
must be added to <code class="varname">newlib_cflags</code> and the <acronym class="acronym">BSP</acronym> must
define the function <code class="function">__getreent</code>, to return the
reentrancy structure for the current thread.
</p></div><div class="sect1" title="5.5. BSP Configuration and Make file;"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_bsp_config_make"></a>5.5.
<acronym class="acronym">BSP</acronym> Configuration and <span class="command"><strong>Make</strong></span> file;
</h2></div></div></div><p>
There is little documentation for the configuration and <span class="command"><strong>make</strong></span> files
for the <acronym class="acronym">BSP</acronym>s. The general guideline is to copy the baseline
versions of these files in the default platform library,
<code class="filename">libnosys</code>, which is based on the minimal
implementations described in <a class="xref" href="#sec_syscalls" title="5.3. Standard System Call Implementations">Section 5.3</a>.
</p><p>
This application note uses the configuration and <span class="command"><strong>make</strong></span> files for
the <span class="application">OpenRISC 1000</span> to illustrate the key principles.
</p><p>
Building the <acronym class="acronym">BSP</acronym> only uses <span class="application">autoconf</span> and <span class="application">autoheader</span>, but not
<span class="application">automake</span>. So there is a <code class="filename">configure.in</code> (or
<code class="filename">configure.ac</code>) and
<code class="filename">Makefile.in</code>, but no
<code class="filename">Makefile.am</code>. After making any changes it is
important to run <span class="application">autoconf</span> and <span class="application">autoheader</span> to regenerate the
<code class="filename">configure</code> script and header files. It will also
need a <code class="filename">aclocal.m4</code> to give the local macro
definitions, which can be regenerated from the main <code class="systemitem">libgloss</code>
<code class="filename">acinclude.m4</code> using <span class="application">aclocal</span>. The command
needed are:
</p><div class="informalfigure"><pre class="programlisting">
aclocal -I ..
autoheader
autoconf
</pre></div><p>
<span class="application">aclocal</span> need only be run the first time the directory is
created. <span class="application">autoheader</span> is only needed if the <acronym class="acronym">BSP</acronym> needs configuration
parameters from the system in a local <code class="filename">config.h</code>
file.
</p><div class="sect2" title="5.5.1. configure.in for the BSP"><div class="titlepage"><div><div><h3 class="title"><a id="id2720649"></a>5.5.1.
<code class="filename">configure.in</code> for the <acronym class="acronym">BSP</acronym>
</h3></div></div></div><p>
The <code class="filename">configure.in</code> for the <span class="application">OpenRISC 1000</span> is closely
based on the version in <code class="filename">libnosys</code>.
</p><p>
The initial declarations just need modifying to change the name of
the package.
</p><div class="informalfigure"><pre class="programlisting">
AC_PREREQ(2.59)
AC_INIT(libor32.a,0.2.0)
AC_CONFIG_HEADER(config.h)
</pre></div><p>
There is then code to print a warning if the user has asked for
shared library support (not available) and to locate the
auxiliary tools for <span class="application">autoconf</span>.
</p><p>
The script makes use of <code class="literal">AC_CANONICAL_SYSTEM</code> to
determine the system type and set appropriate variables. This is
now obsolete, and is replaced by
<code class="literal">AC_CANONICAL_TARGET</code> in the <span class="application">OpenRISC 1000</span> version. The
installed program names may be changed (for example by
<code class="literal">--prefix</code>), so we need
<code class="literal">AC_ARG_PROGRAM</code> and we locate the install
program.
</p><div class="informalfigure"><pre class="programlisting">
AC_CANONICAL_TARGET
AC_ARG_PROGRAM
AC_PROG_INSTALL
</pre></div><p>
The assumption is made that we are using <acronym class="acronym">GNU</acronym> <span class="application">ld</span>, so we define
<code class="literal">HAVE_GNU_LD</code>. The script in
<code class="filename">libnosys</code> does this in an obsolete way, which
is fixed in the <span class="application">OpenRISC 1000</span> script.
</p><div class="informalfigure"><pre class="programlisting">
AC_DEFINE(HAVE_GNU_LD, 1, [Using GNU ld])
</pre></div><p>
The standard script tests the canonical target name to determine
if this is an <acronym class="acronym">ELF</acronym> target. For <span class="application">OpenRISC 1000</span> this is
always the case, so the test can be replaced by a simple
declaration.
</p><div class="informalfigure"><pre class="programlisting">
AC_DEFINE(HAVE_ELF, 1, [Using ELF format])
</pre></div><p>
The script in <code class="filename">libnosys</code> then tests for the
presence of various features. Most of those are not relevant to
<span class="application">OpenRISC 1000</span> so can be left out. However we do need to determine what
the symbol prefix is. We could just define this as being '_', but
instead we let the script work it out, using the standard script's
code.
</p><div class="informalfigure"><pre class="programlisting">
AC_CACHE_CHECK([for symbol prefix], libc_symbol_prefix, [dnl
cat > conftest.c <<\EOF
foo () { }
EOF
libc_symbol_prefix=none
if AC_TRY_COMMAND([${CC-cc} -S conftest.c -o - | fgrep "\$foo" > /dev/null]);
then
libc_symbol_prefix='$'
else
if AC_TRY_COMMAND([${CC-cc} -S conftest.c -o - | fgrep "_foo" > /dev/null]);
then
libc_symbol_prefix=_
fi
fi
rm -f conftest* ])
if test $libc_symbol_prefix != none; then
AC_DEFINE_UNQUOTED(__SYMBOL_PREFIX, "$libc_symbol_prefix", [symbol prefix])
else
AC_DEFINE(__SYMBOL_PREFIX, "", [symbol prefix])
fi
</pre></div><p>
The code to define the various host tools used is
standard. However it will expect to find an
<code class="filename">aclocal.m4</code> file in the directory. This can be
regenerated, or simply copied from the
<code class="filename">libnosys</code> directory. The variable
<code class="varname">host_makefile_frag</code> refers to standard <span class="command"><strong>make</strong></span>
script defining how compilation is carried out for the various
source files.
</p><p>
Finally the new <code class="filename">Makefile</code> can be generated in
a suitably initialized environment.
</p><div class="informalfigure"><pre class="programlisting">
AC_CONFIG_FILES(Makefile,
ac_file=Makefile . ${libgloss_topdir}/config-ml.in,
srcdir=${srcdir}
target=${target}
with_multisubdir=${with_multisubdir}
ac_configure_args="${ac_configure_args} --enable-multilib"
CONFIG_SHELL=${CONFIG_SHELL-/bin/sh}
libgloss_topdir=${libgloss_topdir}
)
AC_OUTPUT
</pre></div></div><div class="sect2" title="5.5.2. Makefile.in for the BSP"><div class="titlepage"><div><div><h3 class="title"><a id="id2720960"></a>5.5.2.
<code class="filename">Makefile.in</code> for the <acronym class="acronym">BSP</acronym>
</h3></div></div></div><p>
The first part of <code class="filename">Makefile.in</code> is just
transferring values from <code class="filename">configure</code> and is
used unchanged. The first potential variation is in multilib
handling. If your <acronym class="acronym">GCC</acronym> implements multilibs, then that may need
to be mirrored in the <acronym class="acronym">BSP</acronym> implementation. If not, then there is
no need to set <code class="literal">MULTIDO</code> and
<code class="literal">MULTICLEAN</code> to <code class="literal">true</code> and these
lines can be removed.
</p><p>
The <code class="filename">Makefile.in</code> in
<code class="filename">libnosys</code> includes an option to use
<span class="emphasis"><em>new</em></span> versions of the loader and
assembler. However for most implementations, the plain tool is all
that is needed, so simple transfer of the configured values is
sufficient.
</p><div class="informalfigure"><pre class="programlisting">
CC = @CC@
AS = @AS@
AR = @AR@
LD = @LD@
RANLIB = @RANLIB@
</pre></div><p>
The main tools will already have been transformed to take account
of any prefix (for example using <span class="command"><strong>or32-elf-gcc</strong></span>
rather than <span class="command"><strong>gcc</strong></span>). However this has not been
done for <span class="command"><strong>objdump</strong></span> and
<span class="command"><strong>objcopy</strong></span>, so these are transformed here.
</p><p>
This is the point at which we define the <acronym class="acronym">BSP</acronym>s to be built. Any
custom flags for the compilation can be added to
<code class="literal">CFLAGS</code> here.
</p><div class="informalfigure"><pre class="programlisting">
CFLAGS = -g
</pre></div><p>
We specify the C start up file(s) and <acronym class="acronym">BSP</acronym>(s) to be built.
</p><div class="informalfigure"><pre class="programlisting">
CRT0 = crt0.o
BSP = libor32.a
BSP_UART = libor32uart.a
OUTPUTS = $(CRT0) $(BSP) $(BSP_UART)
</pre></div><div class="important" title="Important" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Important"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Important]" src="./images/important.png" /></td><th align="left">Important</th></tr><tr><td align="left" valign="top"><p>
It is important to define <code class="literal">OUTPUTS</code>. This is
the complete set of programs and libraries being built. It is
used in the <code class="literal">clean</code> and
<code class="literal">install</code> targets.
</p></td></tr></table></div><p>
For each <acronym class="acronym">BSP</acronym> we specify the object files from which it is
built. For the plain <span class="application">OpenRISC 1000</span> <acronym class="acronym">BSP</acronym> we have:
</p><div class="informalfigure"><pre class="programlisting">
OBJS = _exit.o \
close.o \
environ.o \
execve.o \
fork.o \
fstat.o \
getpid.o \
isatty.o \
kill.o \
link.o \
lseek.o \
open.o \
read.o \
sbrk.o \
stat.o \
times.o \
uart-dummy.o \
unlink.o \
wait.o \
write.o
</pre></div><p>
For the <acronym class="acronym">BSP</acronym> with <acronym class="acronym">UART</acronym> support we use many of the same files,
but also have some different files.
</p><div class="informalfigure"><pre class="programlisting">
UART_OBJS = _exit.o \
close.o \
environ.o \
execve.o \
fork.o \
fstat-uart.o \
getpid.o \
isatty-uart.o \
kill.o \
link.o \
lseek-uart.o \
open.o \
read-uart.o \
sbrk.o \
stat.o \
times.o \
uart.o \
unlink.o \
wait.o \
write-uart.o
</pre></div><p>
At this point, the version of <code class="filename">Makefile.in</code> in
<code class="filename">libnosys</code> specifies explicitly the rules for
compiling object files from C and assembler source. However it is
better to incorporate a standard set of rules, using the
<code class="filename">host_makefile_frag</code> reference from the
configuration.
</p><div class="informalfigure"><pre class="programlisting">
@host_makefile_frag@
</pre></div><p>
This is the point at which to specify the first <span class="command"><strong>make</strong></span> rule to
create the C runtime start up files and <acronym class="acronym">BSP</acronym>s.
</p><div class="informalfigure"><pre class="programlisting">
all: ${CRT0} ${BSP} ${BSP_UART}
</pre></div><p>
The object files (including <code class="filename">crt0.o</code>) will be
built automatically, but we need rules to build the libraries from
them.
</p><div class="informalfigure"><pre class="programlisting">
$(BSP): $(OBJS)
${AR} ${ARFLAGS} $@ $(OBJS)
${RANLIB} $@
$(BSP_UART): $(UART_OBJS)
${AR} ${ARFLAGS} $@ $(UART_OBJS)
${RANLIB} $@
</pre></div><p>
The remainder of <code class="filename">Makefile.in</code> is standard. It
provides rules to clean the build directory, to install the
generated <acronym class="acronym">BSP</acronym>(s) and C start up file(s), and rules to ensure
<code class="filename">configure</code> and <code class="filename">Makefile</code>
are regenerated when necessary.
</p><p>
There also hooks to create, clean and install any documentation
(as <span class="command"><strong>info</strong></span> files), which are empty by default.
</p><p>
Very often these rules are sufficient, so long as all the entities
created have been listed in <code class="literal">OUTPUTS</code>. They
should be modified if necessary.
</p></div></div><div class="sect1" title="5.6. The Default BSP, libnosys"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_libnosys"></a>5.6.
The Default <acronym class="acronym">BSP</acronym>, <code class="systemitem">libnosys</code>
</h2></div></div></div><p>
<code class="systemitem">Newlib</code> also builds a default <acronym class="acronym">BSP</acronym>
<code class="filename">libnosys.a</code>. This can be used with the
<code class="literal">-lnosys</code> flag, and provides a convenient way of
testing that code will link correctly in the absence of a full <acronym class="acronym">BSP</acronym>
</p><p>
The code can be found in the <code class="filename">libnosys</code>
sub-directory of the main <code class="systemitem">libgloss</code> directory.
</p><p>
For completeness, the configuration template file,
<code class="literal">configure.in</code>, in this directory should be updated
for any new target that is defining namespace clean versions of the
functions. Each such system is selected using a
<code class="literal">case</code> statement. The new entry for the <span class="application">OpenRISC 1000</span> is as
follows.
</p><div class="informalfigure"><pre class="programlisting">
or32-*-*)
;;
</pre></div><p>
Having updated the configuration template, run <span class="application">autoconf</span> to
regenerate the <code class="filename">configure</code> script file.
</p></div></div><div class="chapter" title="Chapter 6. Configuring, Building and Installing Newlib and Libgloss"><div class="titlepage"><div><div><h2 class="title"><a id="chap_build_install"></a>Chapter 6.
Configuring, Building and Installing <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</h2></div></div></div><p>
Having made all the changes it is not time to configure, build and
install the system. The examples in this chapter for the <span class="application">OpenRISC 1000</span> assume a
unified source tree in <code class="filename">srcw</code> and a build directory,
<code class="filename">bld-or32</code>, with the installation directory prefix
<code class="filename">/opt/or32-elf-new</code>.
</p><div class="sect1" title="6.1. Configuring Newlib and Libgloss"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="id2721527"></a>6.1.
Configuring <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</h2></div></div></div><p>
<code class="systemitem">Newlib</code> is configured as follows.
</p><div class="informalfigure"><pre class="programlisting">
cd bld-or32
../srcw/configure --target=or32-elf --with-newlib --prefix=/opt/or32-elf-new
</pre></div><div class="note" title="Note" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Note"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Note]" src="./images/note.png" /></td><th align="left">Note</th></tr><tr><td align="left" valign="top"><p>
Other options may be needed on the command line if other <acronym class="acronym">GNU</acronym> tools
are being built. However these are the options relevant to <code class="systemitem">newlib</code>
</p></td></tr></table></div></div><div class="sect1" title="6.2. Building Newlib and Libgloss"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="id2721600"></a>6.2.
Building <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</h2></div></div></div><p>
The system is built using <span class="command"><strong>make</strong></span> from within the
<code class="filename">bld-or32</code> directory.
</p><div class="informalfigure"><pre class="programlisting">
make all-target-newlib
make all-target-libgloss
</pre></div></div><div class="sect1" title="6.3. Testing Newlib and Libgloss"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="id2721651"></a>6.3.
Testing <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</h2></div></div></div><p>
Testing <code class="systemitem">newlib</code> and <code class="systemitem">libgloss</code> requires further configuration. The
details are discussed later in this application note (see <a class="xref" href="#chap_testing" title="Chapter 8. Testing Newlib and Libgloss">Chapter 8</a>). For now this step can be skipped.
</p></div><div class="sect1" title="6.4. Installing Newlib and Libgloss"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="id2721699"></a>6.4.
Installing <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</h2></div></div></div><p>
The system is installed using <span class="command"><strong>make</strong></span> from within the
<code class="filename">bld-or32</code> directory.
</p><div class="informalfigure"><pre class="programlisting">
make install-target-newlib
make install-target-libgloss
</pre></div></div></div><div class="chapter" title="Chapter 7. Modifying the GNU Tool Chain"><div class="titlepage"><div><div><h2 class="title"><a id="id2721746"></a>Chapter 7.
Modifying the <acronym class="acronym">GNU</acronym> Tool Chain
</h2></div></div></div><div class="sect1" title="7.1. Putting Newlib in a Custom Location"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_custom_newlib_loc"></a>7.1.
Putting <code class="systemitem">Newlib</code> in a Custom Location
</h2></div></div></div><p>
Normally <code class="systemitem">newlib</code> will be installed in a standard place with the rest
of the tool chain. Its headers will go in the
<code class="filename">include</code> directory within the target specific
installation directory. The C runtime start up file, the <code class="systemitem">newlib</code>
libraries themselves and <acronym class="acronym">BSP</acronym> libraries will go in the
<code class="filename">lib</code> directory within the target specific
installation directory.
</p><p>
This arrangement ensures that GCC will pick up the headers and
libraries automatically and in the correct sequence.
</p><p>
However if <code class="systemitem">newlib</code> is not the only C library, then this may be
inconvenient. For example the <span class="application">OpenRISC 1000</span> usually uses <code class="systemitem">uClibc</code>, and
only uses <code class="systemitem">newlib</code> when regression testing the <acronym class="acronym">GNU</acronym> tool chain.
</p><p>
The solution is to move the <code class="systemitem">newlib</code> headers and libraries to a custom
location and modify <acronym class="acronym">GCC</acronym> to search there when <code class="systemitem">newlib</code> is being used
(see <a class="xref" href="#sec_changing_gcc" title="7.2. Changes to GCC">Section 7.2</a>).
</p><p>
This is achieved with a simple script at the end of build and
install. For example with the <span class="application">OpenRISC 1000</span> the following command will
suffice, where the prefix used for the entire tool chain build is in
<code class="literal">${install_dir}</code>.
</p><div class="informalfigure"><pre class="programlisting">
mkdir -p ${install_dir}/or32-elf/newlib
rm -rf ${install_dir}/or32-elf/newlib-include
mv ${install_dir}/or32-elf/include ${install_dir}/or32-elf/newlib-include
mv ${install_dir}/or32-elf/lib/*.a ${install_dir}/or32-elf/newlib
mv ${install_dir}/or32-elf/lib/crt0.o ${install_dir}/or32-elf/newlib
</pre></div></div><div class="sect1" title="7.2. Changes to GCC"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_changing_gcc"></a>7.2.
Changes to <acronym class="acronym">GCC</acronym>
</h2></div></div></div><p>
In general <acronym class="acronym">GCC</acronym> will work with <code class="systemitem">newlib</code> with no change. All that is
needed is to include the <acronym class="acronym">BSP</acronym> library on the command line.
</p><p>
However it is convenient to modify <acronym class="acronym">GCC</acronym> so that it picks up the <acronym class="acronym">BSP</acronym>
automatically. This is particularly useful when <code class="systemitem">newlib</code> has been
installed in a custom location (see <a class="xref" href="#sec_custom_newlib_loc" title="7.1. Putting Newlib in a Custom Location">Section 7.1</a>).
</p><p>
This is achieved by adding machine specific options to <acronym class="acronym">GCC</acronym>, and
modifying the Spec definitions to pick up the <code class="systemitem">newlib</code> libraries when
the relevant option is in effect.
</p><p>
All the relevant files are found in the
<code class="filename">gcc/config/<em class="replaceable"><code>target</code></em></code>
directory of <acronym class="acronym">GCC</acronym>. For the 32-bit <span class="application">OpenRISC 1000</span> this is
<code class="filename">gcc/config/or32</code>.
</p><div class="sect2" title="7.2.1. Adding Machine Specific Options for Newlib"><div class="titlepage"><div><div><h3 class="title"><a id="sec_gcc_machine_opts"></a>7.2.1.
Adding Machine Specific Options for <code class="systemitem">Newlib</code>
</h3></div></div></div><p>
Machine specific options are described in the
<code class="filename"><em class="replaceable"><code>target</code></em>.opt</code> file. By
convention machine specific options begin with 'm'.
</p><p>
For the <span class="application">OpenRISC 1000</span> we define two options,
<code class="literal">-mor32-newlib</code> and
<code class="literal">-mor32-newlib-uart</code> for the plain and <acronym class="acronym">UART</acronym>
enabled versions of the <acronym class="acronym">BSP</acronym> respectively.
</p><p>
For each option we provide its name on one line, any parameters on
subsequent lines and a final line of description. In this case the
only parameter is to say that the parameter can only appear in its
positive form (i.e. <code class="literal">--mno-or32-newlib</code> is not
permitted).
</p><div class="informalfigure"><pre class="programlisting">
mor32-newlib
Target RejectNegative
Link with the OR32 newlib library
mor32-newlib-uart
Target RejectNegative
Link with the OR32 newlib UART library
</pre></div><p>
These parameters can then be used elsewhere.
</p></div><div class="sect2" title="7.2.2. Updating Spec Definitions"><div class="titlepage"><div><div><h3 class="title"><a id="sec_gcc_specs"></a>7.2.2.
Updating Spec Definitions
</h3></div></div></div><p>
<acronym class="acronym">GCC</acronym> calls a number of subsidiary programs (the compiler itself,
the assembler, the linker etc). The arguments to these are built up
from the parametrized strings, known as <span class="emphasis"><em>Spec</em></span>
strings.
</p><p>
This application note cannot describe the huge range of possible
parameters. However we will use one example to show what is
possible. The changes are all made to the definitions of the strings
in <code class="filename"><em class="replaceable"><code>target</code></em>.h</code>. In the
case of the <span class="application">OpenRISC 1000</span> this is <code class="filename">or32.h</code>.
</p><p>
We need to make four changes.
</p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>
We need to tell the C preprocessor to look for headers in the
relocated newlib library directory.
</p></li><li class="listitem"><p>
We need to tell the linker to pick up the newlib C runtime
start up file.
</p></li><li class="listitem"><p>
We need to tell the linker where to find the <code class="systemitem">newlib</code> libraries.
</p></li><li class="listitem"><p>
We need to tell the linker to include the <acronym class="acronym">BSP</acronym> library in the
right place.
</p></li></ol></div><div class="sect3" title="The Target Specific Installation Directory"><div class="titlepage"><div><div><h4 class="title"><a id="id2722220"></a>
The Target Specific Installation Directory
</h4></div></div></div><p>
All of these changes will require knowing the location of the
target specific installation directory. Unfortunately there is no
Spec parameter giving this. However we can construct it from two
definitions available when compiling
<acronym class="acronym">GCC</acronym>. <code class="literal">STANDARD_EXEC_PREFIX</code> is the directory
where the <acronym class="acronym">GCC</acronym> executables will be found. Two directories up from
that will be the main prefix directory. The target machine is
specified in <code class="literal">DEFAULT_TARGET_MACHINE</code>. So
concatenating the three strings yields the target specific
directory.
</p><div class="informalfigure"><pre class="programlisting">
STANDARD_EXEC_PREFIX "/../../" DEFAULT_TARGET_MACHINE
</pre></div><p>
The newlib headers are in the subdirectory
<code class="filename">newlib-include</code> and the C runtime start up and
libraries in <code class="filename">newlib</code>.
</p><p>
We define a new string, <code class="literal">TARGET_PREFIX</code> based on
the concatenation.
</p><div class="informalfigure"><pre class="programlisting">
#define CONC_DIR(dir1, dir2) dir1 "/../../" dir2
#define TARGET_PREFIX CONC_DIR (STANDARD_EXEC_PREFIX, DEFAULT_TARGET_MACHINE)
</pre></div><p>
Defined constants cannot be used directly in Spec strings, but we
can make them available by defining the macro
<code class="literal">EXTRA_SPECS</code>.
</p><div class="informalfigure"><pre class="programlisting">
#define EXTRA_SPECS \
{ "target_prefix", TARGET_PREFIX }
</pre></div><p>
The Spec string <code class="literal">target_prefix</code> is now available
to be used in other Spec strings.
</p></div><div class="sect3" title="Specifying the header directory."><div class="titlepage"><div><div><h4 class="title"><a id="id2722338"></a>
Specifying the header directory.
</h4></div></div></div><p>
Additional arguments to the C preprocessor are defined in
<code class="literal">CPP_SPEC</code>. The <code class="systemitem">newlib</code> header directory should
we searched after any user specified header directories (from
<code class="literal">-I</code> arguments) and after the <acronym class="acronym">GCC</acronym> system
headers. So it is specified using the
<code class="literal">-idirafter</code> option.
</p><div class="informalfigure"><pre class="programlisting">
#undef CPP_SPEC
#define CPP_SPEC "%{mor32-newlib*:-idirafter %(target_prefix)/newlib-include}"
</pre></div><p>
This specifies that any option beginning
<code class="literal">-mor32-newlib</code> should be replaced by the string
<code class="literal">-idirafter</code> followed by the
<code class="filename">newlib-incldue</code> subdirectory of the
<code class="literal">target_prefix</code> directory.
</p><p>
So so for example, if we build the <span class="application">OpenRISC 1000</span> <acronym class="acronym">GCC</acronym> with
<code class="literal">--prefix=/opt/or32-elf-new</code>, we would have
<code class="literal">STANDARD_EXEC_PREFIX</code> set to
<code class="filename">/opt/or32-elf-new/lib/gcc</code> and
<code class="literal">DEFAULT_TARGET_MACHINE</code> set to
<code class="filename">or32-elf</code>. The Spec variable
<code class="literal">target_prefix</code> would therefore be
<code class="literal">/opt/or32-elf-new/lib/gcc/../../or32-elf</code> and
thus the C preprocessor would have the following added to its
option list.
</p><div class="informalfigure"><pre class="programlisting">
-idirafter /opt/or32-elf-new/lib/gcc/../../or32-elf/newlib-include"
</pre></div><p>
This substitution only occurs when
<code class="literal">-mor32-newlib</code> or
<code class="literal">-mor32-newlib-uart</code> is specified, which is
exactly the behavior desired.
</p><div class="note" title="Note" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Note"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Note]" src="./images/note.png" /></td><th align="left">Note</th></tr><tr><td align="left" valign="top"><p>
If <code class="systemitem">newlib</code> is not relocated as described in <a class="xref" href="#sec_custom_newlib_loc" title="7.1. Putting Newlib in a Custom Location">Section 7.1</a>, then the headers will be in a
standard location, which <acronym class="acronym">GCC</acronym> will search anyway, so there is
no need to define <code class="literal">CPP_SPEC</code>.
</p></td></tr></table></div></div><div class="sect3" title="Specifying the C Start up File"><div class="titlepage"><div><div><h4 class="title"><a id="id2722528"></a>
Specifying the C Start up File
</h4></div></div></div><p>
<code class="filename">crt0.o</code> should be the first object file or
library specified to the linker. This is covered by
<code class="literal">STARTFILE_SPEC</code>.
</p><p>
This string already has a partial definition, to look for
<code class="filename">crt0.o</code> in a standard place, and to include
the <code class="filename">crtinit.o</code> file from a standard place.
</p><div class="informalfigure"><pre class="programlisting">
#undef STARTFILE_SPEC
#define STARTFILE_SPEC "%{!shared:crt0%s crtinit.o%s}"
</pre></div><p>
So long as <code class="literal">-shared</code> is not specified as an
option, this looks for <code class="literal">crt0.o</code> and
<code class="literal">crtinit.o</code> in standard directories and
substitutes them on the command line (the suffix
<code class="literal">%s</code> indicates that the preceding file should be
searched for in standard directories, and its name expanded to
include the directory name).
</p><p>
This needs changing to indicate that if
<code class="literal">-mor32-newlib</code> or
<code class="literal">-mor32-newlib-uart</code> is specified, then
<code class="literal">crt0.o</code> should be taken from the newlib
directory.
</p><div class="informalfigure"><pre class="programlisting">
#define STARTFILE_SPEC \
"%{!shared:%{mor32-newlib*:%(target_prefix)/newlib/crt0.o} \
%{!mor32-newlib*:crt0.o%s} crtinit.o%s}"
</pre></div><p>
Note that we must also include the case that when neither of the
<code class="systemitem">newlib</code> options is specified, then <code class="filename">crt0.o</code>
will be searched for in standard directories.
</p><div class="note" title="Note" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Note"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Note]" src="./images/note.png" /></td><th align="left">Note</th></tr><tr><td align="left" valign="top"><p>
If <code class="systemitem">newlib</code> is not relocated as described in <a class="xref" href="#sec_custom_newlib_loc" title="7.1. Putting Newlib in a Custom Location">Section 7.1</a>, then
<code class="filename">crt0.o</code> will be in a standard location,
which <acronym class="acronym">GCC</acronym> will search anyway, so there is no need to modify
<code class="literal">STARTFILE_SPEC</code>.
</p></td></tr></table></div></div><div class="sect3" title="Specifying the Newlib library location"><div class="titlepage"><div><div><h4 class="title"><a id="id2722695"></a>
Specifying the <code class="systemitem">Newlib</code> library location
</h4></div></div></div><p>
We need to tell the linker where to look for <code class="systemitem">newlib</code>
libraries. This is achieved in a similar manner to the search for
the headers, but using the <code class="literal">-L</code> option and
<code class="literal">LINK_SPEC</code>.
</p><div class="informalfigure"><pre class="programlisting">
#undef LINK_SPEC
#define LINK_SPEC "%{mor32-newlib*:-L%(target_prefix)/newlib}"
</pre></div><div class="note" title="Note" style="margin-left: 0.5in; margin-right: 0.5in;"><table border="0" summary="Note"><tr><td rowspan="2" align="center" valign="top" width="25"><img alt="[Note]" src="./images/note.png" /></td><th align="left">Note</th></tr><tr><td align="left" valign="top"><p>
If <code class="systemitem">newlib</code> is not relocated as described in <a class="xref" href="#sec_custom_newlib_loc" title="7.1. Putting Newlib in a Custom Location">Section 7.1</a>, then the <code class="systemitem">newlib</code> libraries
will be in a standard location searched by <acronym class="acronym">GCC</acronym>, so there is no
need to specify <code class="literal">LINK_SPEC</code>.
</p></td></tr></table></div></div><div class="sect3" title="Adding a BSP to the link line."><div class="titlepage"><div><div><h4 class="title"><a id="sec_gcc_bsp_location"></a>
Adding a <acronym class="acronym">BSP</acronym> to the link line.
</h4></div></div></div><p>
The libraries searched by <acronym class="acronym">GCC</acronym> are by default specified to be
<code class="literal">-lgcc -lc -lgcc</code>, with variants if
profiling is being used. When a <acronym class="acronym">BSP</acronym> is used, it must be searched
after <code class="systemitem">libc</code>, but that can leave references unresolved, so <code class="systemitem">libc</code>
must be searched again afterward.
</p><p>
The sequence of libraries to be searched between the two searches
of <code class="systemitem">libgcc</code> is given in <code class="literal">LIB_SPEC</code>. It already
has a definition.
</p><div class="informalfigure"><pre class="programlisting">
#define LIB_SPEC "%{!p:%{!pg:-lc}}%{p:-lc_p}%{pg:-lc_p}
</pre></div><p>
This specifies a variant library when profiling is in
place. <code class="systemitem">newlib</code> does not offer profiling support, but it does have
a debugging version of the library (<code class="systemitem">libg</code>).
</p><div class="informalfigure"><pre class="programlisting">
#undef LIB_SPEC
#define LIB_SPEC "%{!mor32-newlib*:%{!p:%{!pg:-lc}}%{p:-lc_p}%{pg:-lc_p}} \
%{mor32-newlib:%{!g:-lc -lor32 -lc} \
%{g:-lg -lor32 -lg}} \
%{mor32-newlib-uart:%{!g:-lc -lor32uart -lc} \
%{g:-lg -lor32uart -lg}}"
</pre></div><p>
This ensures that the correct <acronym class="acronym">BSP</acronym> library will be used,
according the the option selected, and that if
<code class="literal">-g</code> is specified on the command line, the
debugging version of the C library (<code class="systemitem">libg</code>) will be used instead.
</p><p>
Even if the <code class="systemitem">newlib</code> is not relocated as described in <a class="xref" href="#sec_custom_newlib_loc" title="7.1. Putting Newlib in a Custom Location">Section 7.1</a>, then this Spec change is
required in order to ensure the correct libraries are picked up.
</p></div></div></div><div class="sect1" title="7.3. Changes to the GNU Linker"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_linker"></a>7.3.
Changes to the <acronym class="acronym">GNU</acronym> Linker
</h2></div></div></div><p>
In general changes to the linker are not needed. Instead the <acronym class="acronym">BSP</acronym>
should make use of information provided by the standard linker. For
example in the definition of <code class="function">sbrk</code> (see <a class="xref" href="#sec_sbrk" title="5.3.15. Allocate more Heap, sbrk">Section 5.3.15</a>) the code uses the <code class="literal">_end</code> symbol
defined by the linker at the end of the loaded image to be the start
of the heap.
</p></div></div><div class="chapter" title="Chapter 8. Testing Newlib and Libgloss"><div class="titlepage"><div><div><h2 class="title"><a id="chap_testing"></a>Chapter 8.
Testing <code class="systemitem">Newlib</code> and <code class="systemitem">Libgloss</code>
</h2></div></div></div><p>
<code class="systemitem">Newlib</code> and <code class="systemitem">libgloss</code> both come with <span class="application">DejaGnu</span> test infrastructures,
although as noted in <a class="xref" href="#sec_testing_libgloss" title="8.2. Testing Libgloss">Section 8.2</a>, the
<code class="systemitem">libgloss</code> infrastructure is non-functional.
</p><p>
The total number of tests is modest (24 tests in release 1.18.0). In
practice much of the testing is achieved through the <acronym class="acronym">GCC</acronym> test suite
(40,000+ tests) and the <acronym class="acronym">GDB</acronym> test suite (5,000+ tests).
</p><div class="sect1" title="8.1. Testing Newlib"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_testing_newlib"></a>8.1.
Testing <code class="systemitem">Newlib</code>
</h2></div></div></div><p>
Like all tools, <code class="systemitem">newlib</code> can be tested with a <span class="application">DejaGnu</span>
test suite. <span class="application">DejaGnu</span> must be installed on the test machine.
</p><p>
If you already have testing set up for other tools in the <acronym class="acronym">GNU</acronym> tool
chain on your target, then you can skip the remainder of this section,
and just test <code class="systemitem">newlib</code> from the build directory with the following.
</p><div class="informalfigure"><pre class="programlisting">
cd bld-or32
make check-target-newlib
</pre></div><p>
If this is the first time you have tried testing, then you'll need to
set up your system appropriately. Once this is done, you will be able
to test all the <acronym class="acronym">GNU</acronym> tool chain components.
</p><p>
The tests require a target on which to run the tests. This can be a
physical machine, or it can be a simulator for the target
architecture.
</p><p>
The details of the target are provided in an <span class="application">expect</span> board
configuration file. This is referenced from the <span class="application">DejaGnu</span> global
configuration file. The environment variable
<code class="literal">DEJAGNU</code> should point to the global configuration
file.
</p><p>
For the <span class="application">OpenRISC 1000</span> , the global configuration file is in
<code class="filename">site.exp</code> and a subdirectory,
<code class="filename">boards</code> contains
<code class="filename">or32-sim.exp</code>, which is the board configuration
file for the OpenRISC simulator target.
</p><p>
The <code class="filename">site.exp</code> file has two functions. First, it
must add the <code class="filename">boards</code> directory to the list of
board directories to search. Secondly, it must ensure that the target
triplet name is mapped to the name of the board configuration file.
</p><p>
This <code class="filename">site.exp</code> file can be reused for checking
other targets in the <acronym class="acronym">GNU</acronym> tool chain, which may have a different
test suite hierarchy. We cannot therefore just reference the
<code class="filename">boards</code> directory relative to the test
directory. All we know is that it will be in one of the directories
above, and there is no other boards directory in the hierarchy, so we
add all the possible directories. Not elegant, but effective.
</p><div class="informalfigure"><pre class="programlisting">
#Make sure we look in the right place for the board description files
if ![info exists boards_dir] {
set boards_dir {}
}
# Crude way of finding the boards directory
lappend boards_dir "${tool_root_dir}/../boards"
lappend boards_dir "${tool_root_dir}/../../boards"
lappend boards_dir "${tool_root_dir}/../../../boards"
lappend boards_dir "${tool_root_dir}/../../../../boards"
global target_list
case "$target_triplet" in {
{ "or32-*-elf" } {
set target_list { "or32-sim" }
}
}
</pre></div><p>
Within the <code class="filename">boards</code> directory, the board
configuration file, <code class="filename">or32-sim.cfg</code> gives all the
details required for the configuration.
</p><p>
The tool chains supported by this board are specified first. In the
case of the <span class="application">OpenRISC 1000</span> , only one is supported.
</p><div class="informalfigure"><pre class="programlisting">
set_board_info target_install {or32-elf}
</pre></div><p>
We then need to load some generic routines, and the generic board
configuration.
</p><div class="informalfigure"><pre class="programlisting">
load_generic_config "sim"
load_base_board_description "basic-sim"
</pre></div><p>
The default settings assume that a program is executed on the target
by a command named <span class="command"><strong>run</strong></span>, built in a target specific
subdirectory of the top level <code class="filename">sim</code> directory. In
the case of the <span class="application">OpenRISC 1000</span> this directory would be
<code class="filename">sim/or32</code>.
</p><p>
At a minimum, <span class="command"><strong>run</strong></span> takes as argument an executable
to run, and returns the exit code from that executable as its result.
</p><p>
The <code class="filename">sim</code> directory is usually distributed as part
of <acronym class="acronym">GDB</acronym>. Simulators may be derived from <span class="application">CGEN</span> specifications of the
architecture, or by integrating third party simulators. The latter is
the case for the <span class="application">OpenRISC 1000</span> .
</p><p>
The default settings for a target are obtained using the
<code class="literal">setup_sim</code> procedure.
</p><div class="informalfigure"><pre class="programlisting">
setup_sim or32
</pre></div><p>
The remainder of the file is used to configure variations on the
default settings. This is done using the
<code class="literal">set_board_info</code> procedure.
</p><p>
The <span class="application">OpenRISC 1000</span> simulator needs an additional argument, which is a
configuration file for the simulator. We know that file will be in the
<code class="systemitem">libgloss</code> target directory and named <code class="filename">sim.cfg</code>. We
can use the <code class="literal">lookfor_file</code> procedure to search up
from the current source directory to locate the file.
</p><div class="informalfigure"><pre class="programlisting">
set cfg_file [lookfor_file ${srcdir} libgloss/or32/sim.cfg]
set_board_info sim,options "-a \"-f ${cfg_file}\""
</pre></div><p>
A number of helpful procedures make it easy to locate parts of the
tool chain and their default arguments. For the <span class="application">OpenRISC 1000</span> we make one
change, which is to specify <code class="literal">-mor32-newlib</code> for the
linker flags, so that the <code class="systemitem">newlib</code> <acronym class="acronym">BSP</acronym> will be used.
</p><div class="informalfigure"><pre class="programlisting">
process_multilib_options ""
set_board_info compiler "[find_gcc]"
set_board_info cflags "[libgloss_include_flags] [newlib_include_flags]"
set_board_info ldflags "[libgloss_link_flags] -mor32-newlib [newlib_link_flags]"
set_board_info ldscript ""
</pre></div><p>
Not all targets have the same functionality, and the remaining options
specify those limitations. This is a generic board specification, so
some of these apply to testing components other than <code class="systemitem">newlib</code>. The
limitations specified will mean that some tests, which are
inappropriate do not run.
</p><p>
For the <span class="application">OpenRISC 1000</span> we specify that the simulator is fast, that programs it
runs cannot be passed arguments, that it does not support signals
(for testing <acronym class="acronym">GDB</acronym>) and that the maximum stack size is 64KB (for
testing <acronym class="acronym">GCC</acronym>).
</p><div class="informalfigure"><pre class="programlisting">
set_board_info slow_simulator 0
set_board_info noargs 1
set_board_info gdb,nosignals 1
set_board_info gcc,stack_size 65536
</pre></div><p>
We can now set <code class="literal">DEJAGNU</code> to point to the global
configuration directory, change to the build directory and run the
<span class="command"><strong>make</strong></span> command to check newlib.
</p><div class="informalfigure"><pre class="programlisting">
export DEJAGNU=`pwd`/site.exp
cd bld-or32
make check-target-newlib
</pre></div><p>
The good thing is that this set up is generic across all the <acronym class="acronym">GNU</acronym>
tool chain, so all the other tools can be checked in the same way.
</p><div class="sect2" title="8.1.1. Checking Physical Hardware"><div class="titlepage"><div><div><h3 class="title"><a id="id2723637"></a>8.1.1.
Checking Physical Hardware
</h3></div></div></div><p>
The same technique can be used to run the tests against physical
hardware rather than a simulator. The setup of the board
configuration is rather more complicated, with considerable
variation for different arrangements.
</p><p>
The detail is beyond the scope of this application note, but is well
described in Dan Kegel's <span class="emphasis"><em>Crosstool</em></span> project
<a class="xref" href="#ref_crosstool" title="The Crosstool Project,">[2]</a>.
</p></div></div><div class="sect1" title="8.2. Testing Libgloss"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="sec_testing_libgloss"></a>8.2.
Testing <code class="systemitem">Libgloss</code>
</h2></div></div></div><p>
In principle, having set up <code class="systemitem">newlib</code> testing, testing <code class="systemitem">libgloss</code>
should be as simple as:
</p><div class="informalfigure"><pre class="programlisting">
cd bld-or32
make check-target-libgloss
</pre></div><p>
Unfortunately, the current <code class="systemitem">newlib</code> release (at the time of writing
1.18.0) does not implement testing for <code class="systemitem">libgloss</code>. The
<code class="filename">testsuite</code> subdirectory exists, but the code to
configure it is currently commented out in
<code class="filename">configure.in</code>.
</p><p> It should not be difficult to build the infrastructure. However
as noted at the start of this chapter, testing of <code class="systemitem">newlib</code> and
<code class="systemitem">libgloss</code> is as much achieved through <acronym class="acronym">GCC</acronym> and <acronym class="acronym">GDB</acronym> testing as
through the modest number of tests within <code class="systemitem">newlib</code>
</p></div></div><div class="chapter" title="Chapter 9. Summary Checklist"><div class="titlepage"><div><div><h2 class="title"><a id="chap_checklist"></a>Chapter 9.
Summary Checklist
</h2></div></div></div><p>
This summary can be used as a checklist when creating a new port of
<code class="systemitem">newlib</code>. The configuration and build steps are typically encapsulated
in a simple shell script, which builds, tests and installs the entire
<acronym class="acronym">GNU</acronym> tool chain as well as <code class="systemitem">newlib</code> and <code class="systemitem">libgloss</code>
</p><p>
Throughout this checklist, the new target architecture is referred to as
<em class="replaceable"><code>target</code></em>. It is recommended <code class="systemitem">newlib</code> and
<code class="systemitem">libgloss</code> are built as part of a unified source tree including the
<code class="systemitem">newlib</code> distribution (see <a class="xref" href="#sec_unified_source" title="2.1. The Unified Source Tree">Section 2.1</a>).
</p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>
Edit <code class="filename">newlib/configure.host</code> adding entries for
the new target (<a class="xref" href="#sec_adding_target" title="3.3. Adding a new Target to Newlib">Section 3.3</a>).
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
Decide whether to implement reentrant or non-reentrant system
calls and whether to use namespace clean system call names
(<a class="xref" href="#sec_namespace_reent" title="3.2. The C Namespace and Reentrant Functions">Section 3.2</a>).
</p></li></ul></div></li><li class="listitem"><p>
Add a <code class="systemitem">newlib</code> machine subdirectory for the new target,
<code class="filename">newlib/libc/machine/<em class="replaceable"><code>target</code></em></code>
(<a class="xref" href="#sec_machine_dir" title="4.1. The Machine Directory">Section 4.1</a>).
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
Modify <code class="filename">configure.in</code> in
<code class="filename">newlib/libc/machine</code> to configure the new
target subdirectory and run <span class="application">autoconf</span> in
<code class="filename">newlib/libc/machine</code> to regenerate the
<code class="filename">configure</code> script (<a class="xref" href="#sec_machine_dir_reconf" title="4.1.1. Updating the Main Machine Directory Configuration files">Section 4.1.1</a>).
</p></li></ul></div></li><li class="listitem"><p>
Implement <code class="function">setjmp</code> and
<code class="function">longjmp</code> in the target specific machine
directory,
<code class="filename">newlib/libc/machine/<em class="replaceable"><code>target</code></em></code>
(<a class="xref" href="#sec_setjmp_longjmp" title="4.1.2. Implementing the setjmp and longjmp functions.">Section 4.1.2</a>).
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
Copy and modify <code class="filename">Makefile.am</code> and
<code class="filename">configure.in</code> from the <span class="application">fr30</span> directory and
run <span class="application">aclocal</span>, <span class="application">autoconf</span> and <span class="application">automake</span> in
<code class="filename">newlib/libc/machine/<em class="replaceable"><code>target</code></em></code>
to regenerate the <code class="filename">configure</code> script and
Makefile template. (<a class="xref" href="#sec_machine_target_dir_reconf" title="4.1.3. Updating the Target Specific Machine Directory Configuration files">Section 4.1.3</a>).
</p></li></ul></div></li><li class="listitem"><p>
Modify <code class="systemitem">newlib</code> header files (<a class="xref" href="#sec_newlib_headers" title="4.2. Changing Headers">Section 4.2</a>).
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
Add entries in <code class="filename">newlib/libc/include/ieeefp.c</code>
(<a class="xref" href="#sec_fp_header" title="4.2.1. IEEE Floating Point">Section 4.2.1</a>)
</p></li><li class="listitem"><p>
Add entry in in
<code class="filename">newlib/libc/include/setjmp.c</code> (<a class="xref" href="#sec_setjmp_header" title="4.2.2. setjmp Buffer Size">Section 4.2.2</a>)
</p></li><li class="listitem"><p>
Add entries in
<code class="filename">newlib/libc/include/sys/config.h</code> (<a class="xref" href="#sec_newlib_config_header" title="4.2.3. Miscellaneous System Definitions">Section 4.2.3</a>).
</p></li><li class="listitem"><p>
Optionally add other custom headers in
<code class="filename">newlib/libc/machine/target/machine</code> (<a class="xref" href="#sec_other_headers" title="4.2.4. Overriding Other Header Files">Section 4.2.4</a>).
</p></li></ul></div></li><li class="listitem"><p>
Add a <code class="systemitem">libgloss</code> platform directory,
<code class="filename">libgloss/<em class="replaceable"><code>target</code></em></code>
(<a class="xref" href="#sec_platform_dir" title="5.1. The Platform Directory">Section 5.1</a>).
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
Modify <code class="filename">libgloss/configure.in</code> to configure the
platform subdirectory and run <span class="application">autoconf</span> in the
<code class="filename">libgloss</code> directory to regenerate the
<code class="filename">configure</code> script (<a class="xref" href="#sec_platform_dir_config" title="5.1.1. Ensuring the Platform Directory is Configured">Section 5.1.1</a>).
</p></li></ul></div></li><li class="listitem"><p>
Implement the Board Support Package(s) for the target (<a class="xref" href="#chap_libgloss" title="Chapter 5. Modifying libgloss">Chapter 5</a>).
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
Implement the C Runtime start up,
<code class="filename">crt0.o</code> for each <acronym class="acronym">BSP</acronym> (<a class="xref" href="#sec_crt0" title="5.2. The C Runtime Initialization, crt0.o">Section 5.2</a>).
</p></li><li class="listitem"><p>
Implement the environment global variable and 18 system call
functions for each <acronym class="acronym">BSP</acronym> following the convention namespace and
reentrancy conventions specified in
<code class="filename">newlib/configure.host</code> (<a class="xref" href="#sec_syscalls" title="5.3. Standard System Call Implementations">Section 5.3</a> and <a class="xref" href="#sec_reentrant_syscalls" title="5.4. Reentrant System Call Implementations">Section 5.4</a>).
</p></li><li class="listitem"><p>
Create
<code class="filename">libgloss/<em class="replaceable"><code>target</code></em>/Makefile.in</code>
and
<code class="filename">libgloss/<em class="replaceable"><code>target</code></em>/configure.ac</code>,
based on the versions in the
<code class="filename">libgloss/libnosys</code> directory and run
<span class="application">aclocal</span> and <span class="application">autoconf</span> in
<code class="filename">libgloss/<em class="replaceable"><code>target</code></em></code>
(<a class="xref" href="#sec_bsp_config_make" title="5.5. BSP Configuration and Make file;">Section 5.5</a>).
</p></li></ul></div></li><li class="listitem"><p>
If necessary update
<code class="filename">libgloss/libnosys/configure.in</code> to indicate the
target is using namespace clean system calls and run <span class="application">autoconf</span> in
<code class="filename">libgloss/libnosys</code> (<a class="xref" href="#sec_libnosys" title="5.6. The Default BSP, libnosys">Section 5.6</a>).
</p></li><li class="listitem"><p>
Modify <acronym class="acronym">GCC</acronym> for <code class="systemitem">newlib</code> (<a class="xref" href="#sec_changing_gcc" title="7.2. Changes to GCC">Section 7.2</a>).
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
Optionally add target specific option(s) to specify <code class="systemitem">newlib</code>
<acronym class="acronym">BSP</acronym>(s) (<a class="xref" href="#sec_gcc_machine_opts" title="7.2.1. Adding Machine Specific Options for Newlib">Section 7.2.1</a>).
</p></li><li class="listitem"><p>
Optionally specify the location of <code class="systemitem">newlib</code> headers, the <acronym class="acronym">BSP</acronym> C
runtime start up file and the newlib libraries, if they have been
moved from their standard locations and/or names (<a class="xref" href="#sec_gcc_specs" title="7.2.2. Updating Spec Definitions">Section 7.2.2</a>)
</p></li><li class="listitem"><p>
Specify the <code class="systemitem">libgloss</code> <acronym class="acronym">BSP</acronym> library to be linked, ensuring
<code class="function">malloc</code> and <code class="function">free</code> are
linked in if required (<a class="xref" href="#sec_gcc_bsp_location" title="Adding a BSP to the link line.">the section called “
Adding a <acronym class="acronym">BSP</acronym> to the link line.
”</a>).
</p></li></ul></div></li><li class="listitem"><p>
Ensure the linker scripts are suitable for use with <code class="systemitem">newlib</code> (<a class="xref" href="#sec_linker" title="7.3. Changes to the GNU Linker">Section 7.3</a>).
</p></li><li class="listitem"><p>
Configure and build <code class="systemitem">newlib</code> and <code class="systemitem">libgloss</code> (<a class="xref" href="#chap_build_install" title="Chapter 6. Configuring, Building and Installing Newlib and Libgloss">Chapter 6</a>).
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
Optionally move the <code class="systemitem">newlib</code> header directory, libraries, C
start-up and <acronym class="acronym">BSP</acronym>(s) to a custom location (<a class="xref" href="#sec_custom_newlib_loc" title="7.1. Putting Newlib in a Custom Location">Section 7.1</a>).
</p></li><li class="listitem"><p>
Rebuild <acronym class="acronym">GCC</acronym>
</p></li><li class="listitem"><p>
Rebuild <span class="application">ld</span> if any linker scripts have been changed.
</p></li></ul></div></li><li class="listitem"><p>
Test <code class="systemitem">newlib</code> (<a class="xref" href="#sec_testing_newlib" title="8.1. Testing Newlib">Section 8.1</a>).
</p></li><li class="listitem"><p>
Install <code class="systemitem">newlib</code> and <code class="systemitem">libgloss</code> (<a class="xref" href="#chap_build_install" title="Chapter 6. Configuring, Building and Installing Newlib and Libgloss">Chapter 6</a>).
</p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
Reinstall <acronym class="acronym">GCC</acronym>
</p></li><li class="listitem"><p>
Reinstall <span class="application">ld</span> if any linker scripts have been changed.
</p></li></ul></div></li></ol></div><p>
You should now have a working <code class="systemitem">newlib</code> implementation integrated within
your <acronym class="acronym">GNU</acronym> tool chain.
</p></div><div class="glossary" title="Glossary"><div class="titlepage"><div><div><h2 class="title"><a id="id2724638"></a>
Glossary
</h2></div></div></div><dl><dt><a id="id2724644"></a>Application Binary Interface (ABI)</dt><dd xmlns=""><p xmlns="http://www.w3.org/1999/xhtml">
The definition of how registers are used during function call and
return for a particular architecture.
</p></dd><dt><a id="id2724662"></a>big endian</dt><dd xmlns=""><p xmlns="http://www.w3.org/1999/xhtml">
A multi-byte number representation, in which the most significant
byte is placed first (i.e. at the lowest address) in memory.
</p><p>
See also little endian
</p></dd><dt><a id="id2724681"></a>Board Support Package (BSP)</dt><dd xmlns=""><p xmlns="http://www.w3.org/1999/xhtml">
The low level interface between an operating system or library and
the underlying physical platform.
</p></dd><dt><a id="gloss_bss"></a>Block Stated by Symbol (BSS)</dt><dd xmlns=""><p xmlns="http://www.w3.org/1999/xhtml">
Universally known by its acronym (the full name is a historical
relic), this refers to an area of storage used for holding static
variables and initialized to zero.
</p></dd><dt><a id="id2724720"></a>little endian</dt><dd xmlns=""><p xmlns="http://www.w3.org/1999/xhtml">
A multi-byte number representation, in which the least significant
byte is placed first (i.e. at the lowest address) in memory.
</p><p>
See also big endian
</p></dd><dt><a id="id2724738"></a>reentrant</dt><dd xmlns=""><p xmlns="http://www.w3.org/1999/xhtml">
A function which is <span class="emphasis"><em>reentrant</em></span> may be safely
called from another thread of control while an initial thread's flow
of control is still within the function.
</p><p xmlns="http://www.w3.org/1999/xhtml">
In general a function will be reentrant if it changes no static
state.
</p></dd><dt><a id="id2724761"></a>special purpose register (SPR)</dt><dd xmlns=""><p xmlns="http://www.w3.org/1999/xhtml">
A set of up to 2<sup>16</sup> 32-bit registers used
to hold additional information controlling the operation of the
<span class="application">OpenRISC 1000</span>
</p></dd><dt><a id="id2724791"></a>supervision register</dt><dd xmlns=""><p xmlns="http://www.w3.org/1999/xhtml">
An <span class="application">OpenRISC 1000</span> special purpose register holding information about the
most recent test result, whether the processor is in supervisor
mode, and whether certain functions (cache etc) are enabled.
</p><p>
See also special purpose register
</p></dd></dl></div><div class="bibliography" title="References"><div class="titlepage"><div><div><h2 class="title"><a id="id2724819"></a>
References
</h2></div></div></div><div class="bibliomixed" title="The Red Hat Newlib C Library"><a id="ref_libc"></a><p class="bibliomixed" title="The Red Hat Newlib C Library">[1]
<span class="title">
The Red Hat <code class="systemitem">Newlib</code> C Library
</span>
<span class="bibliomisc">
Available at <a class="ulink" href="http://sourceware.org/newlib/libc.html" target="_top">sourceware.org/newlib/libc.html</a>.
</span>
</p></div><div class="bibliomixed" title="The Crosstool Project,"><a id="ref_crosstool"></a><p class="bibliomixed" title="The Crosstool Project,">[2]
<span class="title">
The Crosstool Project,
</span>
<span class="author"><span class="firstname">Dan</span><span class="surname">Kegel.</span></span>
<span class="bibliomisc">
Available at <a class="ulink" href="http://www.kegel.com/crosstool/" target="_top">www.kegel.com/crosstool</a>.
</span>
</p></div></div></div></body></html>
|
