Loading...
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 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 | /*
* This file implements the perfmon subsystem which is used
* to program the IA-64 Performance Monitoring Unit (PMU).
*
* Originaly Written by Ganesh Venkitachalam, IBM Corp.
* Copyright (C) 1999 Ganesh Venkitachalam <venkitac@us.ibm.com>
*
* Modifications by Stephane Eranian, Hewlett-Packard Co.
* Modifications by David Mosberger-Tang, Hewlett-Packard Co.
*
* Copyright (C) 1999-2003 Hewlett Packard Co
* Stephane Eranian <eranian@hpl.hp.com>
* David Mosberger-Tang <davidm@hpl.hp.com>
*/
#include <linux/config.h>
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/interrupt.h>
#include <linux/smp_lock.h>
#include <linux/proc_fs.h>
#include <linux/init.h>
#include <linux/vmalloc.h>
#include <linux/wrapper.h>
#include <linux/mm.h>
#include <linux/sysctl.h>
#include <asm/bitops.h>
#include <asm/errno.h>
#include <asm/page.h>
#include <asm/perfmon.h>
#include <asm/processor.h>
#include <asm/signal.h>
#include <asm/system.h>
#include <asm/uaccess.h>
#include <asm/delay.h> /* for ia64_get_itc() */
#ifdef CONFIG_PERFMON
/*
* For PMUs which rely on the debug registers for some features, you must
* you must enable the following flag to activate the support for
* accessing the registers via the perfmonctl() interface.
*/
#if defined(CONFIG_ITANIUM) || defined(CONFIG_MCKINLEY)
#define PFM_PMU_USES_DBR 1
#endif
/*
* perfmon context states
*/
#define PFM_CTX_DISABLED 0
#define PFM_CTX_ENABLED 1
/*
* Reset register flags
*/
#define PFM_PMD_LONG_RESET 1
#define PFM_PMD_SHORT_RESET 2
/*
* Misc macros and definitions
*/
#define PMU_FIRST_COUNTER 4
#define PMU_MAX_PMCS 256
#define PMU_MAX_PMDS 256
/*
* type of a PMU register (bitmask).
* bitmask structure:
* bit0 : register implemented
* bit1 : end marker
* bit2-3 : reserved
* bit4-7 : register type
* bit8-31: reserved
*/
#define PFM_REG_IMPL 0x1 /* register implemented */
#define PFM_REG_END 0x2 /* end marker */
#define PFM_REG_MONITOR (0x1<<4|PFM_REG_IMPL) /* a PMC with a pmc.pm field only */
#define PFM_REG_COUNTING (0x2<<4|PFM_REG_IMPL) /* a PMC with a pmc.pm AND pmc.oi, a PMD used as a counter */
#define PFM_REG_CONTROL (0x3<<4|PFM_REG_IMPL) /* PMU control register */
#define PFM_REG_CONFIG (0x4<<4|PFM_REG_IMPL) /* refine configuration */
#define PFM_REG_BUFFER (0x5<<4|PFM_REG_IMPL) /* PMD used as buffer */
#define PMC_IS_LAST(i) (pmu_conf.pmc_desc[i].type & PFM_REG_END)
#define PMD_IS_LAST(i) (pmu_conf.pmd_desc[i].type & PFM_REG_END)
#define PFM_IS_DISABLED() pmu_conf.disabled
#define PMC_OVFL_NOTIFY(ctx, i) ((ctx)->ctx_soft_pmds[i].flags & PFM_REGFL_OVFL_NOTIFY)
#define PFM_FL_INHERIT_MASK (PFM_FL_INHERIT_NONE|PFM_FL_INHERIT_ONCE|PFM_FL_INHERIT_ALL)
/* i assume unsigned */
#define PMC_IS_IMPL(i) (i< PMU_MAX_PMCS && (pmu_conf.pmc_desc[i].type & PFM_REG_IMPL))
#define PMD_IS_IMPL(i) (i< PMU_MAX_PMDS && (pmu_conf.pmd_desc[i].type & PFM_REG_IMPL))
/* XXX: these three assume that register i is implemented */
#define PMD_IS_COUNTING(i) (pmu_conf.pmd_desc[i].type == PFM_REG_COUNTING)
#define PMC_IS_COUNTING(i) (pmu_conf.pmc_desc[i].type == PFM_REG_COUNTING)
#define PMC_IS_MONITOR(i) (pmu_conf.pmc_desc[i].type == PFM_REG_MONITOR)
#define PMC_DFL_VAL(i) pmu_conf.pmc_desc[i].default_value
#define PMC_RSVD_MASK(i) pmu_conf.pmc_desc[i].reserved_mask
#define PMD_PMD_DEP(i) pmu_conf.pmd_desc[i].dep_pmd[0]
#define PMC_PMD_DEP(i) pmu_conf.pmc_desc[i].dep_pmd[0]
/* k assume unsigned */
#define IBR_IS_IMPL(k) (k<pmu_conf.num_ibrs)
#define DBR_IS_IMPL(k) (k<pmu_conf.num_dbrs)
#define CTX_IS_ENABLED(c) ((c)->ctx_flags.state == PFM_CTX_ENABLED)
#define CTX_OVFL_NOBLOCK(c) ((c)->ctx_fl_block == 0)
#define CTX_INHERIT_MODE(c) ((c)->ctx_fl_inherit)
#define CTX_HAS_SMPL(c) ((c)->ctx_psb != NULL)
/* XXX: does not support more than 64 PMDs */
#define CTX_USED_PMD(ctx, mask) (ctx)->ctx_used_pmds[0] |= (mask)
#define CTX_IS_USED_PMD(ctx, c) (((ctx)->ctx_used_pmds[0] & (1UL << (c))) != 0UL)
#define CTX_USED_IBR(ctx,n) (ctx)->ctx_used_ibrs[(n)>>6] |= 1UL<< ((n) % 64)
#define CTX_USED_DBR(ctx,n) (ctx)->ctx_used_dbrs[(n)>>6] |= 1UL<< ((n) % 64)
#define CTX_USES_DBREGS(ctx) (((pfm_context_t *)(ctx))->ctx_fl_using_dbreg==1)
#define LOCK_CTX(ctx) spin_lock(&(ctx)->ctx_lock)
#define UNLOCK_CTX(ctx) spin_unlock(&(ctx)->ctx_lock)
#define SET_PMU_OWNER(t) do { pmu_owners[smp_processor_id()].owner = (t); } while(0)
#define PMU_OWNER() pmu_owners[smp_processor_id()].owner
#define LOCK_PFS() spin_lock(&pfm_sessions.pfs_lock)
#define UNLOCK_PFS() spin_unlock(&pfm_sessions.pfs_lock)
#define PFM_REG_RETFLAG_SET(flags, val) do { flags &= ~PFM_REG_RETFL_MASK; flags |= (val); } while(0)
#define PFM_CPUINFO_CLEAR(v) __get_cpu_var(pfm_syst_info) &= ~(v)
#define PFM_CPUINFO_SET(v) __get_cpu_var(pfm_syst_info) |= (v)
#ifdef CONFIG_SMP
#define cpu_is_online(i) (cpu_online_map & (1UL << i))
#else
#define cpu_is_online(i) (i==0)
#endif
/*
* debugging
*/
#define DBprintk(a) \
do { \
if (pfm_sysctl.debug >0) { printk("%s.%d: CPU%d ", __FUNCTION__, __LINE__, smp_processor_id()); printk a; } \
} while (0)
#define DBprintk_ovfl(a) \
do { \
if (pfm_sysctl.debug > 0 && pfm_sysctl.debug_ovfl >0) { printk("%s.%d: CPU%d ", __FUNCTION__, __LINE__, smp_processor_id()); printk a; } \
} while (0)
/*
* Architected PMC structure
*/
typedef struct {
unsigned long pmc_plm:4; /* privilege level mask */
unsigned long pmc_ev:1; /* external visibility */
unsigned long pmc_oi:1; /* overflow interrupt */
unsigned long pmc_pm:1; /* privileged monitor */
unsigned long pmc_ig1:1; /* reserved */
unsigned long pmc_es:8; /* event select */
unsigned long pmc_ig2:48; /* reserved */
} pfm_monitor_t;
/*
* There is one such data structure per perfmon context. It is used to describe the
* sampling buffer. It is to be shared among siblings whereas the pfm_context
* is not.
* Therefore we maintain a refcnt which is incremented on fork().
* This buffer is private to the kernel only the actual sampling buffer
* including its header are exposed to the user. This construct allows us to
* export the buffer read-write, if needed, without worrying about security
* problems.
*/
typedef struct _pfm_smpl_buffer_desc {
spinlock_t psb_lock; /* protection lock */
unsigned long psb_refcnt; /* how many users for the buffer */
int psb_flags; /* bitvector of flags (not yet used) */
void *psb_addr; /* points to location of first entry */
unsigned long psb_entries; /* maximum number of entries */
unsigned long psb_size; /* aligned size of buffer */
unsigned long psb_index; /* next free entry slot XXX: must use the one in buffer */
unsigned long psb_entry_size; /* size of each entry including entry header */
perfmon_smpl_hdr_t *psb_hdr; /* points to sampling buffer header */
struct _pfm_smpl_buffer_desc *psb_next; /* next psb, used for rvfreeing of psb_hdr */
} pfm_smpl_buffer_desc_t;
/*
* psb_flags
*/
#define PSB_HAS_VMA 0x1 /* a virtual mapping for the buffer exists */
#define LOCK_PSB(p) spin_lock(&(p)->psb_lock)
#define UNLOCK_PSB(p) spin_unlock(&(p)->psb_lock)
/*
* 64-bit software counter structure
*/
typedef struct {
u64 val; /* virtual 64bit counter value */
u64 lval; /* last value */
u64 long_reset; /* reset value on sampling overflow */
u64 short_reset;/* reset value on overflow */
u64 reset_pmds[4]; /* which other pmds to reset when this counter overflows */
u64 seed; /* seed for random-number generator */
u64 mask; /* mask for random-number generator */
unsigned int flags; /* notify/do not notify */
} pfm_counter_t;
/*
* perfmon context. One per process, is cloned on fork() depending on
* inheritance flags
*/
typedef struct {
unsigned int state:1; /* 0=disabled, 1=enabled */
unsigned int inherit:2; /* inherit mode */
unsigned int block:1; /* when 1, task will blocked on user notifications */
unsigned int system:1; /* do system wide monitoring */
unsigned int frozen:1; /* pmu must be kept frozen on ctxsw in */
unsigned int protected:1; /* allow access to creator of context only */
unsigned int using_dbreg:1; /* using range restrictions (debug registers) */
unsigned int excl_idle:1; /* exclude idle task in system wide session */
unsigned int trap_reason:2; /* reason for going into pfm_block_ovfl_reset() */
unsigned int reserved:21;
} pfm_context_flags_t;
#define PFM_TRAP_REASON_NONE 0x0 /* default value */
#define PFM_TRAP_REASON_BLOCKSIG 0x1 /* we need to block on overflow and signal user */
#define PFM_TRAP_REASON_SIG 0x2 /* we simply need to signal user */
#define PFM_TRAP_REASON_RESET 0x3 /* we need to reset PMDs */
/*
* perfmon context: encapsulates all the state of a monitoring session
* XXX: probably need to change layout
*/
typedef struct pfm_context {
pfm_smpl_buffer_desc_t *ctx_psb; /* sampling buffer, if any */
unsigned long ctx_smpl_vaddr; /* user level virtual address of smpl buffer */
spinlock_t ctx_lock;
pfm_context_flags_t ctx_flags; /* block/noblock */
struct task_struct *ctx_notify_task; /* who to notify on overflow */
struct task_struct *ctx_owner; /* pid of creator (debug) */
unsigned long ctx_ovfl_regs[4]; /* which registers overflowed (notification) */
unsigned long ctx_smpl_regs[4]; /* which registers to record on overflow */
struct semaphore ctx_restart_sem; /* use for blocking notification mode */
unsigned long ctx_used_pmds[4]; /* bitmask of PMD used */
unsigned long ctx_reload_pmds[4]; /* bitmask of PMD to reload on ctxsw */
unsigned long ctx_used_pmcs[4]; /* bitmask PMC used by context */
unsigned long ctx_reload_pmcs[4]; /* bitmask of PMC to reload on ctxsw */
unsigned long ctx_used_ibrs[4]; /* bitmask of used IBR (speedup ctxsw) */
unsigned long ctx_used_dbrs[4]; /* bitmask of used DBR (speedup ctxsw) */
pfm_counter_t ctx_soft_pmds[IA64_NUM_PMD_REGS]; /* XXX: size should be dynamic */
u64 ctx_saved_psr; /* copy of psr used for lazy ctxsw */
unsigned long ctx_saved_cpus_allowed; /* copy of the task cpus_allowed (system wide) */
unsigned int ctx_cpu; /* CPU used by system wide session */
atomic_t ctx_last_cpu; /* CPU id of current or last CPU used */
} pfm_context_t;
#define ctx_fl_inherit ctx_flags.inherit
#define ctx_fl_block ctx_flags.block
#define ctx_fl_system ctx_flags.system
#define ctx_fl_frozen ctx_flags.frozen
#define ctx_fl_protected ctx_flags.protected
#define ctx_fl_using_dbreg ctx_flags.using_dbreg
#define ctx_fl_excl_idle ctx_flags.excl_idle
#define ctx_fl_trap_reason ctx_flags.trap_reason
/*
* global information about all sessions
* mostly used to synchronize between system wide and per-process
*/
typedef struct {
spinlock_t pfs_lock; /* lock the structure */
unsigned int pfs_task_sessions; /* number of per task sessions */
unsigned int pfs_sys_sessions; /* number of per system wide sessions */
unsigned int pfs_sys_use_dbregs; /* incremented when a system wide session uses debug regs */
unsigned int pfs_ptrace_use_dbregs; /* incremented when a process uses debug regs */
struct task_struct *pfs_sys_session[NR_CPUS]; /* point to task owning a system-wide session */
} pfm_session_t;
/*
* information about a PMC or PMD.
* dep_pmd[]: a bitmask of dependent PMD registers
* dep_pmc[]: a bitmask of dependent PMC registers
*/
typedef struct {
unsigned int type;
int pm_pos;
unsigned long default_value; /* power-on default value */
unsigned long reserved_mask; /* bitmask of reserved bits */
int (*read_check)(struct task_struct *task, unsigned int cnum, unsigned long *val, struct pt_regs *regs);
int (*write_check)(struct task_struct *task, unsigned int cnum, unsigned long *val, struct pt_regs *regs);
unsigned long dep_pmd[4];
unsigned long dep_pmc[4];
} pfm_reg_desc_t;
/* assume cnum is a valid monitor */
#define PMC_PM(cnum, val) (((val) >> (pmu_conf.pmc_desc[cnum].pm_pos)) & 0x1)
#define PMC_WR_FUNC(cnum) (pmu_conf.pmc_desc[cnum].write_check)
#define PMD_WR_FUNC(cnum) (pmu_conf.pmd_desc[cnum].write_check)
#define PMD_RD_FUNC(cnum) (pmu_conf.pmd_desc[cnum].read_check)
/*
* This structure is initialized at boot time and contains
* a description of the PMU main characteristics.
*/
typedef struct {
unsigned int disabled; /* indicates if perfmon is working properly */
unsigned long ovfl_val; /* overflow value for generic counters */
unsigned long impl_pmcs[4]; /* bitmask of implemented PMCS */
unsigned long impl_pmds[4]; /* bitmask of implemented PMDS */
unsigned int num_pmcs; /* number of implemented PMCS */
unsigned int num_pmds; /* number of implemented PMDS */
unsigned int num_ibrs; /* number of implemented IBRS */
unsigned int num_dbrs; /* number of implemented DBRS */
unsigned int num_counters; /* number of PMD/PMC counters */
pfm_reg_desc_t *pmc_desc; /* detailed PMC register dependencies descriptions */
pfm_reg_desc_t *pmd_desc; /* detailed PMD register dependencies descriptions */
} pmu_config_t;
/*
* structure used to pass argument to/from remote CPU
* using IPI to check and possibly save the PMU context on SMP systems.
*
* not used in UP kernels
*/
typedef struct {
struct task_struct *task; /* which task we are interested in */
int retval; /* return value of the call: 0=you can proceed, 1=need to wait for completion */
} pfm_smp_ipi_arg_t;
/*
* perfmon command descriptions
*/
typedef struct {
int (*cmd_func)(struct task_struct *task, pfm_context_t *ctx, void *arg, int count, struct pt_regs *regs);
int cmd_flags;
unsigned int cmd_narg;
size_t cmd_argsize;
} pfm_cmd_desc_t;
#define PFM_CMD_PID 0x1 /* command requires pid argument */
#define PFM_CMD_ARG_READ 0x2 /* command must read argument(s) */
#define PFM_CMD_ARG_RW 0x4 /* command must read/write argument(s) */
#define PFM_CMD_CTX 0x8 /* command needs a perfmon context */
#define PFM_CMD_NOCHK 0x10 /* command does not need to check task's state */
#define PFM_CMD_IDX(cmd) (cmd)
#define PFM_CMD_IS_VALID(cmd) ((PFM_CMD_IDX(cmd) >= 0) && (PFM_CMD_IDX(cmd) < PFM_CMD_COUNT) \
&& pfm_cmd_tab[PFM_CMD_IDX(cmd)].cmd_func != NULL)
#define PFM_CMD_USE_PID(cmd) ((pfm_cmd_tab[PFM_CMD_IDX(cmd)].cmd_flags & PFM_CMD_PID) != 0)
#define PFM_CMD_READ_ARG(cmd) ((pfm_cmd_tab[PFM_CMD_IDX(cmd)].cmd_flags & PFM_CMD_ARG_READ) != 0)
#define PFM_CMD_RW_ARG(cmd) ((pfm_cmd_tab[PFM_CMD_IDX(cmd)].cmd_flags & PFM_CMD_ARG_RW) != 0)
#define PFM_CMD_USE_CTX(cmd) ((pfm_cmd_tab[PFM_CMD_IDX(cmd)].cmd_flags & PFM_CMD_CTX) != 0)
#define PFM_CMD_CHK(cmd) ((pfm_cmd_tab[PFM_CMD_IDX(cmd)].cmd_flags & PFM_CMD_NOCHK) == 0)
#define PFM_CMD_ARG_MANY -1 /* cannot be zero */
#define PFM_CMD_NARG(cmd) (pfm_cmd_tab[PFM_CMD_IDX(cmd)].cmd_narg)
#define PFM_CMD_ARG_SIZE(cmd) (pfm_cmd_tab[PFM_CMD_IDX(cmd)].cmd_argsize)
typedef struct {
int debug; /* turn on/off debugging via syslog */
int debug_ovfl; /* turn on/off debug printk in overflow handler */
int fastctxsw; /* turn on/off fast (unsecure) ctxsw */
} pfm_sysctl_t;
typedef struct {
unsigned long pfm_spurious_ovfl_intr_count; /* keep track of spurious ovfl interrupts */
unsigned long pfm_ovfl_intr_count; /* keep track of ovfl interrupts */
unsigned long pfm_recorded_samples_count;
unsigned long pfm_full_smpl_buffer_count; /* how many times the sampling buffer was full */
char pad[SMP_CACHE_BYTES] ____cacheline_aligned;
} pfm_stats_t;
/*
* perfmon internal variables
*/
static pfm_session_t pfm_sessions; /* global sessions information */
static struct proc_dir_entry *perfmon_dir; /* for debug only */
static pfm_stats_t pfm_stats[NR_CPUS];
static pfm_intr_handler_desc_t *pfm_alternate_intr_handler;
DEFINE_PER_CPU(unsigned long, pfm_syst_info);
/* sysctl() controls */
static pfm_sysctl_t pfm_sysctl;
static ctl_table pfm_ctl_table[]={
{1, "debug", &pfm_sysctl.debug, sizeof(int), 0666, NULL, &proc_dointvec, NULL,},
{2, "debug_ovfl", &pfm_sysctl.debug_ovfl, sizeof(int), 0666, NULL, &proc_dointvec, NULL,},
{3, "fastctxsw", &pfm_sysctl.fastctxsw, sizeof(int), 0600, NULL, &proc_dointvec, NULL,},
{ 0, },
};
static ctl_table pfm_sysctl_dir[] = {
{1, "perfmon", NULL, 0, 0755, pfm_ctl_table, },
{0,},
};
static ctl_table pfm_sysctl_root[] = {
{1, "kernel", NULL, 0, 0755, pfm_sysctl_dir, },
{0,},
};
static struct ctl_table_header *pfm_sysctl_header;
static void pfm_vm_close(struct vm_area_struct * area);
static struct vm_operations_struct pfm_vm_ops={
.close = pfm_vm_close
};
/*
* keep track of task owning the PMU per CPU.
*/
static struct {
struct task_struct *owner;
char pad[SMP_CACHE_BYTES] ____cacheline_aligned;
} pmu_owners[NR_CPUS];
/*
* forward declarations
*/
static void pfm_reset_pmu(struct task_struct *);
static void pfm_lazy_save_regs (struct task_struct *ta);
#if defined(CONFIG_ITANIUM)
#include "perfmon_itanium.h"
#elif defined(CONFIG_MCKINLEY)
#include "perfmon_mckinley.h"
#else
#include "perfmon_generic.h"
#endif
static inline void
pfm_clear_psr_pp(void)
{
__asm__ __volatile__ ("rsm psr.pp;; srlz.i;;"::: "memory");
}
static inline void
pfm_set_psr_pp(void)
{
__asm__ __volatile__ ("ssm psr.pp;; srlz.i;;"::: "memory");
}
static inline void
pfm_clear_psr_up(void)
{
__asm__ __volatile__ ("rum psr.up;; srlz.i;;"::: "memory");
}
static inline void
pfm_set_psr_up(void)
{
__asm__ __volatile__ ("sum psr.up;; srlz.i;;"::: "memory");
}
static inline unsigned long
pfm_get_psr(void)
{
unsigned long tmp;
__asm__ __volatile__ ("mov %0=psr;;": "=r"(tmp) :: "memory");
return tmp;
}
static inline void
pfm_set_psr_l(unsigned long val)
{
__asm__ __volatile__ ("mov psr.l=%0;; srlz.i;;"::"r"(val): "memory");
}
static inline void
pfm_freeze_pmu(void)
{
ia64_set_pmc(0,1UL);
ia64_srlz_d();
}
static inline void
pfm_unfreeze_pmu(void)
{
ia64_set_pmc(0,0UL);
ia64_srlz_d();
}
static inline unsigned long
pfm_read_soft_counter(pfm_context_t *ctx, int i)
{
return ctx->ctx_soft_pmds[i].val + (ia64_get_pmd(i) & pmu_conf.ovfl_val);
}
static inline void
pfm_write_soft_counter(pfm_context_t *ctx, int i, unsigned long val)
{
ctx->ctx_soft_pmds[i].val = val & ~pmu_conf.ovfl_val;
/*
* writing to unimplemented part is ignore, so we do not need to
* mask off top part
*/
ia64_set_pmd(i, val & pmu_conf.ovfl_val);
}
/*
* Generates a unique (per CPU) timestamp
*/
static inline unsigned long
pfm_get_stamp(void)
{
/*
* XXX: must find something more efficient
*/
return ia64_get_itc();
}
/* Here we want the physical address of the memory.
* This is used when initializing the contents of the
* area and marking the pages as reserved.
*/
static inline unsigned long
pfm_kvirt_to_pa(unsigned long adr)
{
__u64 pa = ia64_tpa(adr);
//DBprintk(("kv2pa(%lx-->%lx)\n", adr, pa));
return pa;
}
static void *
pfm_rvmalloc(unsigned long size)
{
void *mem;
unsigned long adr;
size=PAGE_ALIGN(size);
mem=vmalloc(size);
if (mem) {
//printk("perfmon: CPU%d pfm_rvmalloc(%ld)=%p\n", smp_processor_id(), size, mem);
memset(mem, 0, size); /* Clear the ram out, no junk to the user */
adr=(unsigned long) mem;
while (size > 0) {
mem_map_reserve(vmalloc_to_page((void *)adr));
adr+=PAGE_SIZE;
size-=PAGE_SIZE;
}
}
return mem;
}
static void
pfm_rvfree(void *mem, unsigned long size)
{
unsigned long adr;
if (mem) {
adr=(unsigned long) mem;
while ((long) size > 0) {
mem_map_unreserve(vmalloc_to_page((void*)adr));
adr+=PAGE_SIZE;
size-=PAGE_SIZE;
}
vfree(mem);
}
return;
}
/*
* This function gets called from mm/mmap.c:exit_mmap() only when there is a sampling buffer
* attached to the context AND the current task has a mapping for it, i.e., it is the original
* creator of the context.
*
* This function is used to remember the fact that the vma describing the sampling buffer
* has now been removed. It can only be called when no other tasks share the same mm context.
*
*/
static void
pfm_vm_close(struct vm_area_struct *vma)
{
pfm_smpl_buffer_desc_t *psb = (pfm_smpl_buffer_desc_t *)vma->vm_private_data;
if (psb == NULL) {
printk(KERN_DEBUG "perfmon: psb is null in [%d]\n", current->pid);
return;
}
/*
* Add PSB to list of buffers to free on release_thread() when no more users
*
* This call is safe because, once the count is zero is cannot be modified anymore.
* This is not because there is no more user of the mm context, that the sampling
* buffer is not being used anymore outside of this task. In fact, it can still
* be accessed from within the kernel by another task (such as the monitored task).
*
* Therefore, we only move the psb into the list of buffers to free when we know
* nobody else is using it.
* The linked list if independent of the perfmon context, because in the case of
* multi-threaded processes, the last thread may not have been involved with
* monitoring however it will be the one removing the vma and it should therefore
* also remove the sampling buffer. This buffer cannot be removed until the vma
* is removed.
*
* This function cannot remove the buffer from here, because exit_mmap() must first
* complete. Given that there is no other vma related callback in the generic code,
* we have created our own with the linked list of sampling buffers to free. The list
* is part of the thread structure. In release_thread() we check if the list is
* empty. If not we call into perfmon to free the buffer and psb. That is the only
* way to ensure a safe deallocation of the sampling buffer which works when
* the buffer is shared between distinct processes or with multi-threaded programs.
*
* We need to lock the psb because the refcnt test and flag manipulation must
* looked like an atomic operation vis a vis pfm_context_exit()
*/
LOCK_PSB(psb);
if (psb->psb_refcnt == 0) {
psb->psb_next = current->thread.pfm_smpl_buf_list;
current->thread.pfm_smpl_buf_list = psb;
DBprintk(("[%d] add smpl @%p size %lu to smpl_buf_list psb_flags=0x%x\n",
current->pid, psb->psb_hdr, psb->psb_size, psb->psb_flags));
}
DBprintk(("[%d] clearing psb_flags=0x%x smpl @%p size %lu\n",
current->pid, psb->psb_flags, psb->psb_hdr, psb->psb_size));
/*
* decrement the number vma for the buffer
*/
psb->psb_flags &= ~PSB_HAS_VMA;
UNLOCK_PSB(psb);
}
/*
* This function is called from pfm_destroy_context() and also from pfm_inherit()
* to explicitely remove the sampling buffer mapping from the user level address space.
*/
static int
pfm_remove_smpl_mapping(struct task_struct *task)
{
pfm_context_t *ctx = task->thread.pfm_context;
pfm_smpl_buffer_desc_t *psb;
int r;
/*
* some sanity checks first
*/
if (ctx == NULL || task->mm == NULL || ctx->ctx_smpl_vaddr == 0 || ctx->ctx_psb == NULL) {
printk(KERN_DEBUG "perfmon: invalid context mm=%p\n", task->mm);
return -1;
}
psb = ctx->ctx_psb;
down_write(&task->mm->mmap_sem);
r = do_munmap(task->mm, ctx->ctx_smpl_vaddr, psb->psb_size);
up_write(&task->mm->mmap_sem);
if (r !=0) {
printk(KERN_DEBUG "perfmon: pid %d unable to unmap sampling buffer "
"@0x%lx size=%ld\n", task->pid, ctx->ctx_smpl_vaddr, psb->psb_size);
}
DBprintk(("[%d] do_unmap(0x%lx, %ld)=%d refcnt=%lu psb_flags=0x%x\n",
task->pid, ctx->ctx_smpl_vaddr, psb->psb_size, r, psb->psb_refcnt, psb->psb_flags));
return 0;
}
static pfm_context_t *
pfm_context_alloc(void)
{
pfm_context_t *ctx;
/* allocate context descriptor */
ctx = kmalloc(sizeof(pfm_context_t), GFP_KERNEL);
if (ctx) memset(ctx, 0, sizeof(pfm_context_t));
return ctx;
}
static void
pfm_context_free(pfm_context_t *ctx)
{
if (ctx) kfree(ctx);
}
static int
pfm_remap_buffer(struct vm_area_struct *vma, unsigned long buf, unsigned long addr, unsigned long size)
{
unsigned long page;
DBprintk(("CPU%d buf=0x%lx addr=0x%lx size=%ld\n", smp_processor_id(), buf, addr, size));
while (size > 0) {
page = pfm_kvirt_to_pa(buf);
if (remap_page_range(vma, addr, page, PAGE_SIZE, PAGE_READONLY)) return -ENOMEM;
addr += PAGE_SIZE;
buf += PAGE_SIZE;
size -= PAGE_SIZE;
}
return 0;
}
/*
* counts the number of PMDS to save per entry.
* This code is generic enough to accomodate more than 64 PMDS when they become available
*/
static unsigned long
pfm_smpl_entry_size(unsigned long *which, unsigned long size)
{
unsigned long res = 0;
int i;
for (i=0; i < size; i++, which++) res += hweight64(*which);
DBprintk(("weight=%ld\n", res));
return res;
}
/*
* Allocates the sampling buffer and remaps it into caller's address space
*/
static int
pfm_smpl_buffer_alloc(pfm_context_t *ctx, unsigned long *which_pmds, unsigned long entries,
void **user_vaddr)
{
struct mm_struct *mm = current->mm;
struct vm_area_struct *vma = NULL;
unsigned long size, regcount;
void *smpl_buf;
pfm_smpl_buffer_desc_t *psb;
/* note that regcount might be 0, in this case only the header for each
* entry will be recorded.
*/
regcount = pfm_smpl_entry_size(which_pmds, 1);
if ((sizeof(perfmon_smpl_hdr_t)+ entries*sizeof(perfmon_smpl_entry_t)) <= entries) {
DBprintk(("requested entries %lu is too big\n", entries));
return -EINVAL;
}
/*
* 1 buffer hdr and for each entry a header + regcount PMDs to save
*/
size = PAGE_ALIGN( sizeof(perfmon_smpl_hdr_t)
+ entries * (sizeof(perfmon_smpl_entry_t) + regcount*sizeof(u64)));
DBprintk(("sampling buffer size=%lu bytes\n", size));
/*
* check requested size to avoid Denial-of-service attacks
* XXX: may have to refine this test
* Check against address space limit.
*
* if ((mm->total_vm << PAGE_SHIFT) + len> current->rlim[RLIMIT_AS].rlim_cur)
* return -ENOMEM;
*/
if (size > current->rlim[RLIMIT_MEMLOCK].rlim_cur) return -EAGAIN;
/*
* We do the easy to undo allocations first.
*
* pfm_rvmalloc(), clears the buffer, so there is no leak
*/
smpl_buf = pfm_rvmalloc(size);
if (smpl_buf == NULL) {
DBprintk(("Can't allocate sampling buffer\n"));
return -ENOMEM;
}
DBprintk(("smpl_buf @%p\n", smpl_buf));
/* allocate sampling buffer descriptor now */
psb = kmalloc(sizeof(*psb), GFP_KERNEL);
if (psb == NULL) {
DBprintk(("Can't allocate sampling buffer descriptor\n"));
goto error_kmalloc;
}
/* allocate vma */
vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL);
if (!vma) {
DBprintk(("Cannot allocate vma\n"));
goto error_kmem;
}
/*
* partially initialize the vma for the sampling buffer
*
* The VM_DONTCOPY flag is very important as it ensures that the mapping
* will never be inherited for any child process (via fork()) which is always
* what we want.
*/
vma->vm_mm = mm;
vma->vm_flags = VM_READ| VM_MAYREAD |VM_RESERVED|VM_DONTCOPY;
vma->vm_page_prot = PAGE_READONLY; /* XXX may need to change */
vma->vm_ops = &pfm_vm_ops; /* necesarry to get the close() callback */
vma->vm_pgoff = 0;
vma->vm_file = NULL;
vma->vm_private_data = psb; /* information needed by the pfm_vm_close() function */
/*
* Now we have everything we need and we can initialize
* and connect all the data structures
*/
psb->psb_hdr = smpl_buf;
psb->psb_addr = ((char *)smpl_buf)+sizeof(perfmon_smpl_hdr_t); /* first entry */
psb->psb_size = size; /* aligned size */
psb->psb_index = 0;
psb->psb_entries = entries;
psb->psb_refcnt = 1;
psb->psb_flags = PSB_HAS_VMA;
spin_lock_init(&psb->psb_lock);
/*
* XXX: will need to do cacheline alignment to avoid false sharing in SMP mode and
* multitask monitoring.
*/
psb->psb_entry_size = sizeof(perfmon_smpl_entry_t) + regcount*sizeof(u64);
DBprintk(("psb @%p entry_size=%ld hdr=%p addr=%p refcnt=%lu psb_flags=0x%x\n",
(void *)psb,psb->psb_entry_size, (void *)psb->psb_hdr,
(void *)psb->psb_addr, psb->psb_refcnt, psb->psb_flags));
/* initialize some of the fields of user visible buffer header */
psb->psb_hdr->hdr_version = PFM_SMPL_VERSION;
psb->psb_hdr->hdr_entry_size = psb->psb_entry_size;
psb->psb_hdr->hdr_pmds[0] = which_pmds[0];
/*
* Let's do the difficult operations next.
*
* now we atomically find some area in the address space and
* remap the buffer in it.
*/
down_write(¤t->mm->mmap_sem);
/* find some free area in address space, must have mmap sem held */
vma->vm_start = get_unmapped_area(NULL, 0, size, 0, MAP_PRIVATE|MAP_ANONYMOUS);
if (vma->vm_start == 0UL) {
DBprintk(("Cannot find unmapped area for size %ld\n", size));
up_write(¤t->mm->mmap_sem);
goto error;
}
vma->vm_end = vma->vm_start + size;
DBprintk(("entries=%ld aligned size=%ld, unmapped @0x%lx\n", entries, size, vma->vm_start));
/* can only be applied to current, need to have the mm semaphore held when called */
if (pfm_remap_buffer(vma, (unsigned long)smpl_buf, vma->vm_start, size)) {
DBprintk(("Can't remap buffer\n"));
up_write(¤t->mm->mmap_sem);
goto error;
}
/*
* now insert the vma in the vm list for the process, must be
* done with mmap lock held
*/
insert_vm_struct(mm, vma);
mm->total_vm += size >> PAGE_SHIFT;
up_write(¤t->mm->mmap_sem);
/* store which PMDS to record */
ctx->ctx_smpl_regs[0] = which_pmds[0];
/* link to perfmon context */
ctx->ctx_psb = psb;
/*
* keep track of user level virtual address
*/
ctx->ctx_smpl_vaddr = *(unsigned long *)user_vaddr = vma->vm_start;
return 0;
error:
kmem_cache_free(vm_area_cachep, vma);
error_kmem:
kfree(psb);
error_kmalloc:
pfm_rvfree(smpl_buf, size);
return -ENOMEM;
}
static int
pfm_reserve_session(struct task_struct *task, int is_syswide, unsigned long cpu_mask)
{
unsigned long m, undo_mask;
unsigned int n, i;
/*
* validy checks on cpu_mask have been done upstream
*/
LOCK_PFS();
if (is_syswide) {
/*
* cannot mix system wide and per-task sessions
*/
if (pfm_sessions.pfs_task_sessions > 0UL) {
DBprintk(("system wide not possible, %u conflicting task_sessions\n",
pfm_sessions.pfs_task_sessions));
goto abort;
}
m = cpu_mask; undo_mask = 0UL; n = 0;
DBprintk(("cpu_mask=0x%lx\n", cpu_mask));
for(i=0; m; i++, m>>=1) {
if ((m & 0x1) == 0UL) continue;
if (pfm_sessions.pfs_sys_session[i]) goto undo;
DBprintk(("reserving CPU%d currently on CPU%d\n", i, smp_processor_id()));
pfm_sessions.pfs_sys_session[i] = task;
undo_mask |= 1UL << i;
n++;
}
pfm_sessions.pfs_sys_sessions += n;
} else {
if (pfm_sessions.pfs_sys_sessions) goto abort;
pfm_sessions.pfs_task_sessions++;
}
DBprintk(("task_sessions=%u sys_session[%d]=%d",
pfm_sessions.pfs_task_sessions,
smp_processor_id(), pfm_sessions.pfs_sys_session[smp_processor_id()] ? 1 : 0));
UNLOCK_PFS();
return 0;
undo:
DBprintk(("system wide not possible, conflicting session [%d] on CPU%d\n",
pfm_sessions.pfs_sys_session[i]->pid, i));
for(i=0; undo_mask; i++, undo_mask >>=1) {
pfm_sessions.pfs_sys_session[i] = NULL;
}
abort:
UNLOCK_PFS();
return -EBUSY;
}
static int
pfm_unreserve_session(struct task_struct *task, int is_syswide, unsigned long cpu_mask)
{
pfm_context_t *ctx;
unsigned long m;
unsigned int n, i;
ctx = task ? task->thread.pfm_context : NULL;
/*
* validy checks on cpu_mask have been done upstream
*/
LOCK_PFS();
DBprintk(("[%d] sys_sessions=%u task_sessions=%u dbregs=%u syswide=%d cpu_mask=0x%lx\n",
task->pid,
pfm_sessions.pfs_sys_sessions,
pfm_sessions.pfs_task_sessions,
pfm_sessions.pfs_sys_use_dbregs,
is_syswide,
cpu_mask));
if (is_syswide) {
m = cpu_mask; n = 0;
for(i=0; m; i++, m>>=1) {
if ((m & 0x1) == 0UL) continue;
pfm_sessions.pfs_sys_session[i] = NULL;
n++;
}
/*
* would not work with perfmon+more than one bit in cpu_mask
*/
if (ctx && ctx->ctx_fl_using_dbreg) {
if (pfm_sessions.pfs_sys_use_dbregs == 0) {
printk(KERN_DEBUG "perfmon: invalid release for [%d] "
"sys_use_dbregs=0\n", task->pid);
} else {
pfm_sessions.pfs_sys_use_dbregs--;
}
}
pfm_sessions.pfs_sys_sessions -= n;
DBprintk(("CPU%d sys_sessions=%u\n",
smp_processor_id(), pfm_sessions.pfs_sys_sessions));
} else {
pfm_sessions.pfs_task_sessions--;
DBprintk(("[%d] task_sessions=%u\n",
task->pid, pfm_sessions.pfs_task_sessions));
}
UNLOCK_PFS();
return 0;
}
/*
* XXX: do something better here
*/
static int
pfm_bad_permissions(struct task_struct *task)
{
/* stolen from bad_signal() */
return (current->session != task->session)
&& (current->euid ^ task->suid) && (current->euid ^ task->uid)
&& (current->uid ^ task->suid) && (current->uid ^ task->uid);
}
static int
pfx_is_sane(struct task_struct *task, pfarg_context_t *pfx)
{
unsigned long smpl_pmds = pfx->ctx_smpl_regs[0];
int ctx_flags;
int cpu;
/* valid signal */
/* cannot send to process 1, 0 means do not notify */
if (pfx->ctx_notify_pid == 1) {
DBprintk(("invalid notify_pid %d\n", pfx->ctx_notify_pid));
return -EINVAL;
}
ctx_flags = pfx->ctx_flags;
if ((ctx_flags & PFM_FL_INHERIT_MASK) == (PFM_FL_INHERIT_ONCE|PFM_FL_INHERIT_ALL)) {
DBprintk(("invalid inherit mask 0x%x\n",ctx_flags & PFM_FL_INHERIT_MASK));
return -EINVAL;
}
if (ctx_flags & PFM_FL_SYSTEM_WIDE) {
DBprintk(("cpu_mask=0x%lx\n", pfx->ctx_cpu_mask));
/*
* cannot block in this mode
*/
if (ctx_flags & PFM_FL_NOTIFY_BLOCK) {
DBprintk(("cannot use blocking mode when in system wide monitoring\n"));
return -EINVAL;
}
/*
* must only have one bit set in the CPU mask
*/
if (hweight64(pfx->ctx_cpu_mask) != 1UL) {
DBprintk(("invalid CPU mask specified\n"));
return -EINVAL;
}
/*
* and it must be a valid CPU
*/
cpu = ffz(~pfx->ctx_cpu_mask);
if (cpu_is_online(cpu) == 0) {
DBprintk(("CPU%d is not online\n", cpu));
return -EINVAL;
}
/*
* check for pre-existing pinning, if conflicting reject
*/
if (task->cpus_allowed != ~0UL && (task->cpus_allowed & (1UL<<cpu)) == 0) {
DBprintk(("[%d] pinned on 0x%lx, mask for CPU%d \n", task->pid,
task->cpus_allowed, cpu));
return -EINVAL;
}
} else {
/*
* must provide a target for the signal in blocking mode even when
* no counter is configured with PFM_FL_REG_OVFL_NOTIFY
*/
if ((ctx_flags & PFM_FL_NOTIFY_BLOCK) && pfx->ctx_notify_pid == 0) {
DBprintk(("must have notify_pid when blocking for [%d]\n", task->pid));
return -EINVAL;
}
#if 0
if ((ctx_flags & PFM_FL_NOTIFY_BLOCK) && pfx->ctx_notify_pid == task->pid) {
DBprintk(("cannot notify self when blocking for [%d]\n", task->pid));
return -EINVAL;
}
#endif
}
/* verify validity of smpl_regs */
if ((smpl_pmds & pmu_conf.impl_pmds[0]) != smpl_pmds) {
DBprintk(("invalid smpl_regs 0x%lx\n", smpl_pmds));
return -EINVAL;
}
/* probably more to add here */
return 0;
}
static int
pfm_context_create(struct task_struct *task, pfm_context_t *ctx, void *req, int count,
struct pt_regs *regs)
{
pfarg_context_t tmp;
void *uaddr = NULL;
int ret;
int ctx_flags;
pid_t notify_pid;
/* a context has already been defined */
if (ctx) return -EBUSY;
/*
* not yet supported
*/
if (task != current) return -EINVAL;
if (__copy_from_user(&tmp, req, sizeof(tmp))) return -EFAULT;
ret = pfx_is_sane(task, &tmp);
if (ret < 0) return ret;
ctx_flags = tmp.ctx_flags;
ret = pfm_reserve_session(task, ctx_flags & PFM_FL_SYSTEM_WIDE, tmp.ctx_cpu_mask);
if (ret) goto abort;
ret = -ENOMEM;
ctx = pfm_context_alloc();
if (!ctx) goto error;
/* record the creator (important for inheritance) */
ctx->ctx_owner = current;
notify_pid = tmp.ctx_notify_pid;
spin_lock_init(&ctx->ctx_lock);
if (notify_pid == current->pid) {
ctx->ctx_notify_task = current;
task->thread.pfm_context = ctx;
} else if (notify_pid!=0) {
struct task_struct *notify_task;
read_lock(&tasklist_lock);
notify_task = find_task_by_pid(notify_pid);
if (notify_task) {
ret = -EPERM;
/*
* check if we can send this task a signal
*/
if (pfm_bad_permissions(notify_task)) {
read_unlock(&tasklist_lock);
goto buffer_error;
}
/*
* make visible
* must be done inside critical section
*
* if the initialization does not go through it is still
* okay because child will do the scan for nothing which
* won't hurt.
*/
task->thread.pfm_context = ctx;
/*
* will cause task to check on exit for monitored
* processes that would notify it. see release_thread()
* Note: the scan MUST be done in release thread, once the
* task has been detached from the tasklist otherwise you are
* exposed to race conditions.
*/
atomic_add(1, &ctx->ctx_notify_task->thread.pfm_notifiers_check);
ctx->ctx_notify_task = notify_task;
}
read_unlock(&tasklist_lock);
}
/*
* notification process does not exist
*/
if (notify_pid != 0 && ctx->ctx_notify_task == NULL) {
ret = -EINVAL;
goto buffer_error;
}
if (tmp.ctx_smpl_entries) {
DBprintk(("sampling entries=%lu\n",tmp.ctx_smpl_entries));
ret = pfm_smpl_buffer_alloc(ctx, tmp.ctx_smpl_regs,
tmp.ctx_smpl_entries, &uaddr);
if (ret<0) goto buffer_error;
tmp.ctx_smpl_vaddr = uaddr;
}
/* initialization of context's flags */
ctx->ctx_fl_inherit = ctx_flags & PFM_FL_INHERIT_MASK;
ctx->ctx_fl_block = (ctx_flags & PFM_FL_NOTIFY_BLOCK) ? 1 : 0;
ctx->ctx_fl_system = (ctx_flags & PFM_FL_SYSTEM_WIDE) ? 1: 0;
ctx->ctx_fl_excl_idle = (ctx_flags & PFM_FL_EXCL_IDLE) ? 1: 0;
ctx->ctx_fl_frozen = 0;
ctx->ctx_fl_trap_reason = PFM_TRAP_REASON_NONE;
/*
* setting this flag to 0 here means, that the creator or the task that the
* context is being attached are granted access. Given that a context can only
* be created for the calling process this, in effect only allows the creator
* to access the context. See pfm_protect() for more.
*/
ctx->ctx_fl_protected = 0;
/* for system wide mode only (only 1 bit set) */
ctx->ctx_cpu = ffz(~tmp.ctx_cpu_mask);
atomic_set(&ctx->ctx_last_cpu,-1); /* SMP only, means no CPU */
sema_init(&ctx->ctx_restart_sem, 0); /* init this semaphore to locked */
if (__copy_to_user(req, &tmp, sizeof(tmp))) {
ret = -EFAULT;
goto buffer_error;
}
DBprintk(("context=%p, pid=%d notify_task=%p\n",
(void *)ctx, task->pid, ctx->ctx_notify_task));
DBprintk(("context=%p, pid=%d flags=0x%x inherit=%d block=%d system=%d excl_idle=%d\n",
(void *)ctx, task->pid, ctx_flags, ctx->ctx_fl_inherit,
ctx->ctx_fl_block, ctx->ctx_fl_system, ctx->ctx_fl_excl_idle));
/*
* when no notification is required, we can make this visible at the last moment
*/
if (notify_pid == 0) task->thread.pfm_context = ctx;
/*
* pin task to CPU and force reschedule on exit to ensure
* that when back to user level the task runs on the designated
* CPU.
*/
if (ctx->ctx_fl_system) {
ctx->ctx_saved_cpus_allowed = task->cpus_allowed;
set_cpus_allowed(task, tmp.ctx_cpu_mask);
DBprintk(("[%d] rescheduled allowed=0x%lx\n", task->pid, task->cpus_allowed));
}
return 0;
buffer_error:
pfm_context_free(ctx);
error:
pfm_unreserve_session(task, ctx_flags & PFM_FL_SYSTEM_WIDE , tmp.ctx_cpu_mask);
abort:
/* make sure we don't leave anything behind */
task->thread.pfm_context = NULL;
return ret;
}
static inline unsigned long
pfm_new_counter_value (pfm_counter_t *reg, int is_long_reset)
{
unsigned long val = is_long_reset ? reg->long_reset : reg->short_reset;
unsigned long new_seed, old_seed = reg->seed, mask = reg->mask;
extern unsigned long carta_random32 (unsigned long seed);
if (reg->flags & PFM_REGFL_RANDOM) {
new_seed = carta_random32(old_seed);
val -= (old_seed & mask); /* counter values are negative numbers! */
if ((mask >> 32) != 0)
/* construct a full 64-bit random value: */
new_seed |= carta_random32(old_seed >> 32) << 32;
reg->seed = new_seed;
}
reg->lval = val;
return val;
}
static void
pfm_reset_regs(pfm_context_t *ctx, unsigned long *ovfl_regs, int flag)
{
unsigned long mask = ovfl_regs[0];
unsigned long reset_others = 0UL;
unsigned long val;
int i, is_long_reset = (flag == PFM_PMD_LONG_RESET);
/*
* now restore reset value on sampling overflowed counters
*/
mask >>= PMU_FIRST_COUNTER;
for(i = PMU_FIRST_COUNTER; mask; i++, mask >>= 1) {
if (mask & 0x1) {
val = pfm_new_counter_value(ctx->ctx_soft_pmds + i, is_long_reset);
reset_others |= ctx->ctx_soft_pmds[i].reset_pmds[0];
DBprintk_ovfl(("[%d] %s reset soft_pmd[%d]=%lx\n", current->pid,
is_long_reset ? "long" : "short", i, val));
/* upper part is ignored on rval */
pfm_write_soft_counter(ctx, i, val);
}
}
/*
* Now take care of resetting the other registers
*/
for(i = 0; reset_others; i++, reset_others >>= 1) {
if ((reset_others & 0x1) == 0) continue;
val = pfm_new_counter_value(ctx->ctx_soft_pmds + i, is_long_reset);
if (PMD_IS_COUNTING(i)) {
pfm_write_soft_counter(ctx, i, val);
} else {
ia64_set_pmd(i, val);
}
DBprintk_ovfl(("[%d] %s reset_others pmd[%d]=%lx\n", current->pid,
is_long_reset ? "long" : "short", i, val));
}
ia64_srlz_d();
}
static int
pfm_write_pmcs(struct task_struct *task, pfm_context_t *ctx, void *arg, int count, struct pt_regs *regs)
{
struct thread_struct *th = &task->thread;
pfarg_reg_t tmp, *req = (pfarg_reg_t *)arg;
unsigned long value, reset_pmds;
unsigned int cnum, reg_flags, flags;
int i;
int ret = -EINVAL;
/* we don't quite support this right now */
if (task != current) return -EINVAL;
if (!CTX_IS_ENABLED(ctx)) return -EINVAL;
/* XXX: ctx locking may be required here */
for (i = 0; i < count; i++, req++) {
if (__copy_from_user(&tmp, req, sizeof(tmp))) return -EFAULT;
cnum = tmp.reg_num;
reg_flags = tmp.reg_flags;
value = tmp.reg_value;
reset_pmds = tmp.reg_reset_pmds[0];
flags = 0;
/*
* we reject all non implemented PMC as well
* as attempts to modify PMC[0-3] which are used
* as status registers by the PMU
*/
if (!PMC_IS_IMPL(cnum) || cnum < 4) {
DBprintk(("pmc[%u] is unimplemented or invalid\n", cnum));
goto error;
}
/*
* A PMC used to configure monitors must be:
* - system-wide session: privileged monitor
* - per-task : user monitor
* any other configuration is rejected.
*/
if (PMC_IS_MONITOR(cnum) || PMC_IS_COUNTING(cnum)) {
DBprintk(("pmc[%u].pm=%ld\n", cnum, PMC_PM(cnum, value)));
if (ctx->ctx_fl_system ^ PMC_PM(cnum, value)) {
DBprintk(("pmc_pm=%ld fl_system=%d\n", PMC_PM(cnum, value), ctx->ctx_fl_system));
goto error;
}
}
if (PMC_IS_COUNTING(cnum)) {
pfm_monitor_t *p = (pfm_monitor_t *)&value;
/*
* enforce generation of overflow interrupt. Necessary on all
* CPUs.
*/
p->pmc_oi = 1;
if (reg_flags & PFM_REGFL_OVFL_NOTIFY) {
/*
* must have a target for the signal
*/
if (ctx->ctx_notify_task == NULL) {
DBprintk(("cannot set ovfl_notify: no notify_task\n"));
goto error;
}
flags |= PFM_REGFL_OVFL_NOTIFY;
}
if (reg_flags & PFM_REGFL_RANDOM) flags |= PFM_REGFL_RANDOM;
/* verify validity of reset_pmds */
if ((reset_pmds & pmu_conf.impl_pmds[0]) != reset_pmds) {
DBprintk(("invalid reset_pmds 0x%lx for pmc%u\n", reset_pmds, cnum));
goto error;
}
} else if (reg_flags & (PFM_REGFL_OVFL_NOTIFY|PFM_REGFL_RANDOM)) {
DBprintk(("cannot set ovfl_notify or random on pmc%u\n", cnum));
goto error;
}
/*
* execute write checker, if any
*/
if (PMC_WR_FUNC(cnum)) {
ret = PMC_WR_FUNC(cnum)(task, cnum, &value, regs);
if (ret) goto error;
ret = -EINVAL;
}
/*
* no error on this register
*/
PFM_REG_RETFLAG_SET(tmp.reg_flags, 0);
/*
* update register return value, abort all if problem during copy.
* we only modify the reg_flags field. no check mode is fine because
* access has been verified upfront in sys_perfmonctl().
*
* If this fails, then the software state is not modified
*/
if (__put_user(tmp.reg_flags, &req->reg_flags)) return -EFAULT;
/*
* Now we commit the changes to the software state
*/
/*
* full flag update each time a register is programmed
*/
ctx->ctx_soft_pmds[cnum].flags = flags;
if (PMC_IS_COUNTING(cnum)) {
ctx->ctx_soft_pmds[cnum].reset_pmds[0] = reset_pmds;
/* mark all PMDS to be accessed as used */
CTX_USED_PMD(ctx, reset_pmds);
}
/*
* Needed in case the user does not initialize the equivalent
* PMD. Clearing is done in reset_pmu() so there is no possible
* leak here.
*/
CTX_USED_PMD(ctx, pmu_conf.pmc_desc[cnum].dep_pmd[0]);
/*
* keep copy the pmc, used for register reload
*/
th->pmc[cnum] = value;
ia64_set_pmc(cnum, value);
DBprintk(("[%d] pmc[%u]=0x%lx flags=0x%x used_pmds=0x%lx\n",
task->pid, cnum, value,
ctx->ctx_soft_pmds[cnum].flags,
ctx->ctx_used_pmds[0]));
}
return 0;
error:
PFM_REG_RETFLAG_SET(tmp.reg_flags, PFM_REG_RETFL_EINVAL);
if (__put_user(tmp.reg_flags, &req->reg_flags)) ret = -EFAULT;
DBprintk(("[%d] pmc[%u]=0x%lx error %d\n", task->pid, cnum, value, ret));
return ret;
}
static int
pfm_write_pmds(struct task_struct *task, pfm_context_t *ctx, void *arg, int count, struct pt_regs *regs)
{
pfarg_reg_t tmp, *req = (pfarg_reg_t *)arg;
unsigned long value, hw_value;
unsigned int cnum;
int i;
int ret = -EINVAL;
/* we don't quite support this right now */
if (task != current) return -EINVAL;
/*
* Cannot do anything before PMU is enabled
*/
if (!CTX_IS_ENABLED(ctx)) return -EINVAL;
preempt_disable();
/* XXX: ctx locking may be required here */
for (i = 0; i < count; i++, req++) {
if (__copy_from_user(&tmp, req, sizeof(tmp))) return -EFAULT;
cnum = tmp.reg_num;
value = tmp.reg_value;
if (!PMD_IS_IMPL(cnum)) {
DBprintk(("pmd[%u] is unimplemented or invalid\n", cnum));
goto abort_mission;
}
/*
* execute write checker, if any
*/
if (PMD_WR_FUNC(cnum)) {
unsigned long v = value;
ret = PMD_WR_FUNC(cnum)(task, cnum, &v, regs);
if (ret) goto abort_mission;
value = v;
ret = -EINVAL;
}
hw_value = value;
/*
* no error on this register
*/
PFM_REG_RETFLAG_SET(tmp.reg_flags, 0);
if (__put_user(tmp.reg_flags, &req->reg_flags)) return -EFAULT;
/*
* now commit changes to software state
*/
/* update virtualized (64bits) counter */
if (PMD_IS_COUNTING(cnum)) {
ctx->ctx_soft_pmds[cnum].lval = value;
ctx->ctx_soft_pmds[cnum].val = value & ~pmu_conf.ovfl_val;
hw_value = value & pmu_conf.ovfl_val;
ctx->ctx_soft_pmds[cnum].long_reset = tmp.reg_long_reset;
ctx->ctx_soft_pmds[cnum].short_reset = tmp.reg_short_reset;
ctx->ctx_soft_pmds[cnum].seed = tmp.reg_random_seed;
ctx->ctx_soft_pmds[cnum].mask = tmp.reg_random_mask;
}
/* keep track of what we use */
CTX_USED_PMD(ctx, pmu_conf.pmd_desc[(cnum)].dep_pmd[0]);
/* mark this register as used as well */
CTX_USED_PMD(ctx, RDEP(cnum));
/* writes to unimplemented part is ignored, so this is safe */
ia64_set_pmd(cnum, hw_value);
/* to go away */
ia64_srlz_d();
DBprintk(("[%d] pmd[%u]: value=0x%lx hw_value=0x%lx soft_pmd=0x%lx short_reset=0x%lx "
"long_reset=0x%lx hw_pmd=%lx notify=%c used_pmds=0x%lx reset_pmds=0x%lx\n",
task->pid, cnum,
value, hw_value,
ctx->ctx_soft_pmds[cnum].val,
ctx->ctx_soft_pmds[cnum].short_reset,
ctx->ctx_soft_pmds[cnum].long_reset,
ia64_get_pmd(cnum) & pmu_conf.ovfl_val,
PMC_OVFL_NOTIFY(ctx, cnum) ? 'Y':'N',
ctx->ctx_used_pmds[0],
ctx->ctx_soft_pmds[cnum].reset_pmds[0]));
}
preempt_enable();
return 0;
abort_mission:
preempt_enable();
/*
* for now, we have only one possibility for error
*/
PFM_REG_RETFLAG_SET(tmp.reg_flags, PFM_REG_RETFL_EINVAL);
/*
* we change the return value to EFAULT in case we cannot write register return code.
* The caller first must correct this error, then a resubmission of the request will
* eventually yield the EINVAL.
*/
if (__put_user(tmp.reg_flags, &req->reg_flags)) ret = -EFAULT;
DBprintk(("[%d] pmc[%u]=0x%lx ret %d\n", task->pid, cnum, value, ret));
return ret;
}
static int
pfm_read_pmds(struct task_struct *task, pfm_context_t *ctx, void *arg, int count, struct pt_regs *regs)
{
struct thread_struct *th = &task->thread;
unsigned long val, lval;
pfarg_reg_t *req = (pfarg_reg_t *)arg;
unsigned int cnum, reg_flags = 0;
int i, ret = 0;
#if __GNUC__ < 3
int foo;
#endif
if (!CTX_IS_ENABLED(ctx)) return -EINVAL;
/*
* XXX: MUST MAKE SURE WE DON"T HAVE ANY PENDING OVERFLOW BEFORE READING
* This is required when the monitoring has been stoppped by user or kernel.
* If it is still going on, then that's fine because we a re not guaranteed
* to return an accurate value in this case.
*/
/* XXX: ctx locking may be required here */
DBprintk(("ctx_last_cpu=%d for [%d]\n", atomic_read(&ctx->ctx_last_cpu), task->pid));
for (i = 0; i < count; i++, req++) {
int me;
#if __GNUC__ < 3
foo = __get_user(cnum, &req->reg_num);
if (foo) return -EFAULT;
foo = __get_user(reg_flags, &req->reg_flags);
if (foo) return -EFAULT;
#else
if (__get_user(cnum, &req->reg_num)) return -EFAULT;
if (__get_user(reg_flags, &req->reg_flags)) return -EFAULT;
#endif
lval = 0UL;
if (!PMD_IS_IMPL(cnum)) goto abort_mission;
/*
* we can only read the register that we use. That includes
* the one we explicitely initialize AND the one we want included
* in the sampling buffer (smpl_regs).
*
* Having this restriction allows optimization in the ctxsw routine
* without compromising security (leaks)
*/
if (!CTX_IS_USED_PMD(ctx, cnum)) goto abort_mission;
/*
* If the task is not the current one, then we check if the
* PMU state is still in the local live register due to lazy ctxsw.
* If true, then we read directly from the registers.
*/
me = get_cpu();
if (atomic_read(&ctx->ctx_last_cpu) == me){
ia64_srlz_d();
val = ia64_get_pmd(cnum);
DBprintk(("reading pmd[%u]=0x%lx from hw\n", cnum, val));
} else {
val = th->pmd[cnum];
}
if (PMD_IS_COUNTING(cnum)) {
/*
* XXX: need to check for overflow
*/
val &= pmu_conf.ovfl_val;
val += ctx->ctx_soft_pmds[cnum].val;
lval = ctx->ctx_soft_pmds[cnum].lval;
}
/*
* execute read checker, if any
*/
if (PMD_RD_FUNC(cnum)) {
unsigned long v = val;
ret = PMD_RD_FUNC(cnum)(task, cnum, &v, regs);
val = v;
}
PFM_REG_RETFLAG_SET(reg_flags, ret);
put_cpu();
DBprintk(("read pmd[%u] ret=%d value=0x%lx pmc=0x%lx\n",
cnum, ret, val, ia64_get_pmc(cnum)));
/*
* update register return value, abort all if problem during copy.
* we only modify the reg_flags field. no check mode is fine because
* access has been verified upfront in sys_perfmonctl().
*/
if (__put_user(cnum, &req->reg_num)) return -EFAULT;
if (__put_user(val, &req->reg_value)) return -EFAULT;
if (__put_user(reg_flags, &req->reg_flags)) return -EFAULT;
if (__put_user(lval, &req->reg_last_reset_value)) return -EFAULT;
}
return 0;
abort_mission:
PFM_REG_RETFLAG_SET(reg_flags, PFM_REG_RETFL_EINVAL);
/*
* XXX: if this fails, we stick with the original failure, flag not updated!
*/
__put_user(reg_flags, &req->reg_flags);
return -EINVAL;
}
#ifdef PFM_PMU_USES_DBR
/*
* Only call this function when a process it trying to
* write the debug registers (reading is always allowed)
*/
int
pfm_use_debug_registers(struct task_struct *task)
{
pfm_context_t *ctx = task->thread.pfm_context;
int ret = 0;
DBprintk(("called for [%d]\n", task->pid));
/*
* do it only once
*/
if (task->thread.flags & IA64_THREAD_DBG_VALID) return 0;
/*
* Even on SMP, we do not need to use an atomic here because
* the only way in is via ptrace() and this is possible only when the
* process is stopped. Even in the case where the ctxsw out is not totally
* completed by the time we come here, there is no way the 'stopped' process
* could be in the middle of fiddling with the pfm_write_ibr_dbr() routine.
* So this is always safe.
*/
if (ctx && ctx->ctx_fl_using_dbreg == 1) return -1;
LOCK_PFS();
/*
* We cannot allow setting breakpoints when system wide monitoring
* sessions are using the debug registers.
*/
if (pfm_sessions.pfs_sys_use_dbregs> 0)
ret = -1;
else
pfm_sessions.pfs_ptrace_use_dbregs++;
DBprintk(("ptrace_use_dbregs=%u sys_use_dbregs=%u by [%d] ret = %d\n",
pfm_sessions.pfs_ptrace_use_dbregs,
pfm_sessions.pfs_sys_use_dbregs,
task->pid, ret));
UNLOCK_PFS();
return ret;
}
/*
* This function is called for every task that exits with the
* IA64_THREAD_DBG_VALID set. This indicates a task which was
* able to use the debug registers for debugging purposes via
* ptrace(). Therefore we know it was not using them for
* perfmormance monitoring, so we only decrement the number
* of "ptraced" debug register users to keep the count up to date
*/
int
pfm_release_debug_registers(struct task_struct *task)
{
int ret;
LOCK_PFS();
if (pfm_sessions.pfs_ptrace_use_dbregs == 0) {
printk(KERN_DEBUG "perfmon: invalid release for [%d] ptrace_use_dbregs=0\n",
task->pid);
ret = -1;
} else {
pfm_sessions.pfs_ptrace_use_dbregs--;
ret = 0;
}
UNLOCK_PFS();
return ret;
}
#else /* PFM_PMU_USES_DBR is true */
/*
* in case, the PMU does not use the debug registers, these two functions are nops.
* The first function is called from arch/ia64/kernel/ptrace.c.
* The second function is called from arch/ia64/kernel/process.c.
*/
int
pfm_use_debug_registers(struct task_struct *task)
{
return 0;
}
int
pfm_release_debug_registers(struct task_struct *task)
{
return 0;
}
#endif /* PFM_PMU_USES_DBR */
static int
pfm_restart(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
void *sem = &ctx->ctx_restart_sem;
/*
* Cannot do anything before PMU is enabled
*/
if (!CTX_IS_ENABLED(ctx)) return -EINVAL;
if (task == current) {
DBprintk(("restarting self %d frozen=%d ovfl_regs=0x%lx\n",
task->pid,
ctx->ctx_fl_frozen,
ctx->ctx_ovfl_regs[0]));
preempt_disable();
pfm_reset_regs(ctx, ctx->ctx_ovfl_regs, PFM_PMD_LONG_RESET);
ctx->ctx_ovfl_regs[0] = 0UL;
/*
* We ignore block/don't block because we never block
* for a self-monitoring process.
*/
ctx->ctx_fl_frozen = 0;
if (CTX_HAS_SMPL(ctx)) {
ctx->ctx_psb->psb_hdr->hdr_count = 0;
ctx->ctx_psb->psb_index = 0;
}
/* simply unfreeze */
pfm_unfreeze_pmu();
preempt_enable();
return 0;
}
/* restart on another task */
/*
* if blocking, then post the semaphore.
* if non-blocking, then we ensure that the task will go into
* pfm_overflow_must_block() before returning to user mode.
* We cannot explicitely reset another task, it MUST always
* be done by the task itself. This works for system wide because
* the tool that is controlling the session is doing "self-monitoring".
*
* XXX: what if the task never goes back to user?
*
*/
if (CTX_OVFL_NOBLOCK(ctx) == 0) {
DBprintk(("unblocking %d \n", task->pid));
up(sem);
} else {
task->thread.pfm_ovfl_block_reset = 1;
}
#if 0
/*
* in case of non blocking mode, then it's just a matter of
* of reseting the sampling buffer (if any) index. The PMU
* is already active.
*/
/*
* must reset the header count first
*/
if (CTX_HAS_SMPL(ctx)) {
DBprintk(("resetting sampling indexes for %d \n", task->pid));
ctx->ctx_psb->psb_hdr->hdr_count = 0;
ctx->ctx_psb->psb_index = 0;
}
#endif
return 0;
}
static int
pfm_stop(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
/* we don't quite support this right now */
if (task != current) return -EINVAL;
/*
* Cannot do anything before PMU is enabled
*/
if (!CTX_IS_ENABLED(ctx)) return -EINVAL;
DBprintk(("[%d] fl_system=%d owner=%p current=%p\n",
current->pid,
ctx->ctx_fl_system, PMU_OWNER(),
current));
preempt_disable();
/* simply stop monitoring but not the PMU */
if (ctx->ctx_fl_system) {
/* disable dcr pp */
ia64_set_dcr(ia64_get_dcr() & ~IA64_DCR_PP);
/* stop monitoring */
pfm_clear_psr_pp();
ia64_srlz_i();
PFM_CPUINFO_CLEAR(PFM_CPUINFO_DCR_PP);
ia64_psr(regs)->pp = 0;
} else {
/* stop monitoring */
pfm_clear_psr_up();
ia64_srlz_i();
/*
* clear user level psr.up
*/
ia64_psr(regs)->up = 0;
}
preempt_enable();
return 0;
}
static int
pfm_disable(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
/* we don't quite support this right now */
if (task != current) return -EINVAL;
if (!CTX_IS_ENABLED(ctx)) return -EINVAL;
preempt_disable();
/*
* stop monitoring, freeze PMU, and save state in context
* this call will clear IA64_THREAD_PM_VALID for per-task sessions.
*/
pfm_flush_regs(task);
if (ctx->ctx_fl_system) {
ia64_psr(regs)->pp = 0;
} else {
ia64_psr(regs)->up = 0;
}
/*
* goes back to default behavior: no user level control
* no need to change live psr.sp because useless at the kernel level
*/
ia64_psr(regs)->sp = 1;
DBprintk(("enabling psr.sp for [%d]\n", current->pid));
ctx->ctx_flags.state = PFM_CTX_DISABLED;
preempt_enable();
return 0;
}
static int
pfm_context_destroy(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
/* we don't quite support this right now */
if (task != current) return -EINVAL;
/*
* if context was never enabled, then there is not much
* to do
*/
if (!CTX_IS_ENABLED(ctx)) goto skipped_stop;
/*
* Disable context: stop monitoring, flush regs to software state (useless here),
* and freeze PMU
*
* The IA64_THREAD_PM_VALID is cleared by pfm_flush_regs() called from pfm_disable()
*/
pfm_disable(task, ctx, arg, count, regs);
if (ctx->ctx_fl_system) {
ia64_psr(regs)->pp = 0;
} else {
ia64_psr(regs)->up = 0;
}
skipped_stop:
/*
* remove sampling buffer mapping, if any
*/
if (ctx->ctx_smpl_vaddr) {
pfm_remove_smpl_mapping(task);
ctx->ctx_smpl_vaddr = 0UL;
}
/* now free context and related state */
pfm_context_exit(task);
return 0;
}
/*
* does nothing at the moment
*/
static int
pfm_context_unprotect(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
return 0;
}
static int
pfm_protect_context(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
DBprintk(("context from [%d] is protected\n", task->pid));
/*
* from now on, only the creator of the context has access to it
*/
ctx->ctx_fl_protected = 1;
/*
* reinforce secure monitoring: cannot toggle psr.up
*/
ia64_psr(regs)->sp = 1;
return 0;
}
static int
pfm_debug(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
unsigned int mode = *(unsigned int *)arg;
pfm_sysctl.debug = mode == 0 ? 0 : 1;
printk(KERN_INFO "perfmon debugging %s\n", pfm_sysctl.debug ? "on" : "off");
return 0;
}
#ifdef PFM_PMU_USES_DBR
typedef struct {
unsigned long ibr_mask:56;
unsigned long ibr_plm:4;
unsigned long ibr_ig:3;
unsigned long ibr_x:1;
} ibr_mask_reg_t;
typedef struct {
unsigned long dbr_mask:56;
unsigned long dbr_plm:4;
unsigned long dbr_ig:2;
unsigned long dbr_w:1;
unsigned long dbr_r:1;
} dbr_mask_reg_t;
typedef union {
unsigned long val;
ibr_mask_reg_t ibr;
dbr_mask_reg_t dbr;
} dbreg_t;
static int
pfm_write_ibr_dbr(int mode, struct task_struct *task, void *arg, int count, struct pt_regs *regs)
{
struct thread_struct *thread = &task->thread;
pfm_context_t *ctx = task->thread.pfm_context;
pfarg_dbreg_t tmp, *req = (pfarg_dbreg_t *)arg;
dbreg_t dbreg;
unsigned int rnum;
int first_time;
int i, ret = 0;
/*
* we do not need to check for ipsr.db because we do clear ibr.x, dbr.r, and dbr.w
* ensuring that no real breakpoint can be installed via this call.
*/
first_time = ctx->ctx_fl_using_dbreg == 0;
/*
* check for debug registers in system wide mode
*
*/
LOCK_PFS();
if (ctx->ctx_fl_system && first_time) {
if (pfm_sessions.pfs_ptrace_use_dbregs)
ret = -EBUSY;
else
pfm_sessions.pfs_sys_use_dbregs++;
}
UNLOCK_PFS();
if (ret != 0) return ret;
if (ctx->ctx_fl_system) {
/* we mark ourselves as owner of the debug registers */
ctx->ctx_fl_using_dbreg = 1;
DBprintk(("system-wide setting fl_using_dbreg for [%d]\n", task->pid));
} else if (first_time) {
ret= -EBUSY;
if ((thread->flags & IA64_THREAD_DBG_VALID) != 0) {
DBprintk(("debug registers already in use for [%d]\n", task->pid));
goto abort_mission;
}
/* we mark ourselves as owner of the debug registers */
ctx->ctx_fl_using_dbreg = 1;
DBprintk(("setting fl_using_dbreg for [%d]\n", task->pid));
/*
* Given debug registers cannot be used for both debugging
* and performance monitoring at the same time, we reuse
* the storage area to save and restore the registers on ctxsw.
*/
memset(task->thread.dbr, 0, sizeof(task->thread.dbr));
memset(task->thread.ibr, 0, sizeof(task->thread.ibr));
}
if (first_time) {
DBprintk(("[%d] clearing ibrs,dbrs\n", task->pid));
/*
* clear hardware registers to make sure we don't
* pick up stale state.
*
* for a system wide session, we do not use
* thread.dbr, thread.ibr because this process
* never leaves the current CPU and the state
* is shared by all processes running on it
*/
for (i=0; i < pmu_conf.num_ibrs; i++) {
ia64_set_ibr(i, 0UL);
}
ia64_srlz_i();
for (i=0; i < pmu_conf.num_dbrs; i++) {
ia64_set_dbr(i, 0UL);
}
ia64_srlz_d();
}
ret = -EFAULT;
/*
* Now install the values into the registers
*/
for (i = 0; i < count; i++, req++) {
if (__copy_from_user(&tmp, req, sizeof(tmp))) goto abort_mission;
rnum = tmp.dbreg_num;
dbreg.val = tmp.dbreg_value;
ret = -EINVAL;
if ((mode == 0 && !IBR_IS_IMPL(rnum)) || ((mode == 1) && !DBR_IS_IMPL(rnum))) {
DBprintk(("invalid register %u val=0x%lx mode=%d i=%d count=%d\n",
rnum, dbreg.val, mode, i, count));
goto abort_mission;
}
/*
* make sure we do not install enabled breakpoint
*/
if (rnum & 0x1) {
if (mode == 0)
dbreg.ibr.ibr_x = 0;
else
dbreg.dbr.dbr_r = dbreg.dbr.dbr_w = 0;
}
/*
* clear return flags and copy back to user
*
* XXX: fix once EAGAIN is implemented
*/
ret = -EFAULT;
PFM_REG_RETFLAG_SET(tmp.dbreg_flags, 0);
if (__copy_to_user(req, &tmp, sizeof(tmp))) goto abort_mission;
/*
* Debug registers, just like PMC, can only be modified
* by a kernel call. Moreover, perfmon() access to those
* registers are centralized in this routine. The hardware
* does not modify the value of these registers, therefore,
* if we save them as they are written, we can avoid having
* to save them on context switch out. This is made possible
* by the fact that when perfmon uses debug registers, ptrace()
* won't be able to modify them concurrently.
*/
if (mode == 0) {
CTX_USED_IBR(ctx, rnum);
ia64_set_ibr(rnum, dbreg.val);
ia64_srlz_i();
thread->ibr[rnum] = dbreg.val;
DBprintk(("write ibr%u=0x%lx used_ibrs=0x%lx\n", rnum, dbreg.val, ctx->ctx_used_ibrs[0]));
} else {
CTX_USED_DBR(ctx, rnum);
ia64_set_dbr(rnum, dbreg.val);
ia64_srlz_d();
thread->dbr[rnum] = dbreg.val;
DBprintk(("write dbr%u=0x%lx used_dbrs=0x%lx\n", rnum, dbreg.val, ctx->ctx_used_dbrs[0]));
}
}
return 0;
abort_mission:
/*
* in case it was our first attempt, we undo the global modifications
*/
if (first_time) {
LOCK_PFS();
if (ctx->ctx_fl_system) {
pfm_sessions.pfs_sys_use_dbregs--;
}
UNLOCK_PFS();
ctx->ctx_fl_using_dbreg = 0;
}
/*
* install error return flag
*/
if (ret != -EFAULT) {
/*
* XXX: for now we can only come here on EINVAL
*/
PFM_REG_RETFLAG_SET(tmp.dbreg_flags, PFM_REG_RETFL_EINVAL);
if (__put_user(tmp.dbreg_flags, &req->dbreg_flags)) ret = -EFAULT;
}
return ret;
}
static int
pfm_write_ibrs(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
/* we don't quite support this right now */
if (task != current) return -EINVAL;
if (!CTX_IS_ENABLED(ctx)) return -EINVAL;
return pfm_write_ibr_dbr(0, task, arg, count, regs);
}
static int
pfm_write_dbrs(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
/* we don't quite support this right now */
if (task != current) return -EINVAL;
if (!CTX_IS_ENABLED(ctx)) return -EINVAL;
return pfm_write_ibr_dbr(1, task, arg, count, regs);
}
#endif /* PFM_PMU_USES_DBR */
static int
pfm_get_features(struct task_struct *task, pfm_context_t *ctx, void *arg, int count, struct pt_regs *regs)
{
pfarg_features_t tmp;
memset(&tmp, 0, sizeof(tmp));
tmp.ft_version = PFM_VERSION;
tmp.ft_smpl_version = PFM_SMPL_VERSION;
if (__copy_to_user(arg, &tmp, sizeof(tmp))) return -EFAULT;
return 0;
}
static int
pfm_start(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
/* we don't quite support this right now */
if (task != current) return -EINVAL;
/*
* Cannot do anything before PMU is enabled
*/
if (!CTX_IS_ENABLED(ctx)) return -EINVAL;
DBprintk(("[%d] fl_system=%d owner=%p current=%p\n",
current->pid,
ctx->ctx_fl_system, PMU_OWNER(),
current));
if (PMU_OWNER() != task) {
printk(KERN_DEBUG "perfmon: pfm_start task [%d] not pmu owner\n", task->pid);
return -EINVAL;
}
preempt_disable();
if (ctx->ctx_fl_system) {
PFM_CPUINFO_SET(PFM_CPUINFO_DCR_PP);
/* set user level psr.pp */
ia64_psr(regs)->pp = 1;
/* start monitoring at kernel level */
pfm_set_psr_pp();
/* enable dcr pp */
ia64_set_dcr(ia64_get_dcr()|IA64_DCR_PP);
ia64_srlz_i();
} else {
if ((task->thread.flags & IA64_THREAD_PM_VALID) == 0) {
preempt_enable();
printk(KERN_DEBUG "perfmon: pfm_start task flag not set for [%d]\n",
task->pid);
return -EINVAL;
}
/* set user level psr.up */
ia64_psr(regs)->up = 1;
/* start monitoring at kernel level */
pfm_set_psr_up();
ia64_srlz_i();
}
preempt_enable();
return 0;
}
static int
pfm_enable(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
int me;
/* we don't quite support this right now */
if (task != current) return -EINVAL;
me = get_cpu(); /* make sure we're not migrated or preempted */
if (ctx->ctx_fl_system == 0 && PMU_OWNER() && PMU_OWNER() != current)
pfm_lazy_save_regs(PMU_OWNER());
/* reset all registers to stable quiet state */
pfm_reset_pmu(task);
/* make sure nothing starts */
if (ctx->ctx_fl_system) {
ia64_psr(regs)->pp = 0;
ia64_psr(regs)->up = 0; /* just to make sure! */
/* make sure monitoring is stopped */
pfm_clear_psr_pp();
ia64_srlz_i();
PFM_CPUINFO_CLEAR(PFM_CPUINFO_DCR_PP);
PFM_CPUINFO_SET(PFM_CPUINFO_SYST_WIDE);
if (ctx->ctx_fl_excl_idle) PFM_CPUINFO_SET(PFM_CPUINFO_EXCL_IDLE);
} else {
/*
* needed in case the task was a passive task during
* a system wide session and now wants to have its own
* session
*/
ia64_psr(regs)->pp = 0; /* just to make sure! */
ia64_psr(regs)->up = 0;
/* make sure monitoring is stopped */
pfm_clear_psr_up();
ia64_srlz_i();
DBprintk(("clearing psr.sp for [%d]\n", current->pid));
/* allow user level control */
ia64_psr(regs)->sp = 0;
/* PMU state will be saved/restored on ctxsw */
task->thread.flags |= IA64_THREAD_PM_VALID;
}
SET_PMU_OWNER(task);
ctx->ctx_flags.state = PFM_CTX_ENABLED;
atomic_set(&ctx->ctx_last_cpu, me);
/* simply unfreeze */
pfm_unfreeze_pmu();
put_cpu();
return 0;
}
static int
pfm_get_pmc_reset(struct task_struct *task, pfm_context_t *ctx, void *arg, int count,
struct pt_regs *regs)
{
pfarg_reg_t tmp, *req = (pfarg_reg_t *)arg;
unsigned int cnum;
int i, ret = -EINVAL;
for (i = 0; i < count; i++, req++) {
if (__copy_from_user(&tmp, req, sizeof(tmp))) return -EFAULT;
cnum = tmp.reg_num;
if (!PMC_IS_IMPL(cnum)) goto abort_mission;
tmp.reg_value = PMC_DFL_VAL(cnum);
PFM_REG_RETFLAG_SET(tmp.reg_flags, 0);
DBprintk(("pmc_reset_val pmc[%u]=0x%lx\n", cnum, tmp.reg_value));
if (__copy_to_user(req, &tmp, sizeof(tmp))) return -EFAULT;
}
return 0;
abort_mission:
PFM_REG_RETFLAG_SET(tmp.reg_flags, PFM_REG_RETFL_EINVAL);
if (__copy_to_user(req, &tmp, sizeof(tmp))) ret = -EFAULT;
return ret;
}
/*
* functions MUST be listed in the increasing order of their index (see permfon.h)
*/
static pfm_cmd_desc_t pfm_cmd_tab[]={
/* 0 */{ NULL, 0, 0, 0}, /* not used */
/* 1 */{ pfm_write_pmcs, PFM_CMD_PID|PFM_CMD_CTX|PFM_CMD_ARG_RW, PFM_CMD_ARG_MANY, sizeof(pfarg_reg_t)},
/* 2 */{ pfm_write_pmds, PFM_CMD_PID|PFM_CMD_CTX|PFM_CMD_ARG_RW, PFM_CMD_ARG_MANY, sizeof(pfarg_reg_t)},
/* 3 */{ pfm_read_pmds,PFM_CMD_PID|PFM_CMD_CTX|PFM_CMD_ARG_RW, PFM_CMD_ARG_MANY, sizeof(pfarg_reg_t)},
/* 4 */{ pfm_stop, PFM_CMD_PID|PFM_CMD_CTX, 0, 0},
/* 5 */{ pfm_start, PFM_CMD_PID|PFM_CMD_CTX, 0, 0},
/* 6 */{ pfm_enable, PFM_CMD_PID|PFM_CMD_CTX, 0, 0},
/* 7 */{ pfm_disable, PFM_CMD_PID|PFM_CMD_CTX, 0, 0},
/* 8 */{ pfm_context_create, PFM_CMD_PID|PFM_CMD_ARG_RW, 1, sizeof(pfarg_context_t)},
/* 9 */{ pfm_context_destroy, PFM_CMD_PID|PFM_CMD_CTX, 0, 0},
/* 10 */{ pfm_restart, PFM_CMD_PID|PFM_CMD_CTX|PFM_CMD_NOCHK, 0, 0},
/* 11 */{ pfm_protect_context, PFM_CMD_PID|PFM_CMD_CTX, 0, 0},
/* 12 */{ pfm_get_features, PFM_CMD_ARG_RW, 0, 0},
/* 13 */{ pfm_debug, 0, 1, sizeof(unsigned int)},
/* 14 */{ pfm_context_unprotect, PFM_CMD_PID|PFM_CMD_CTX, 0, 0},
/* 15 */{ pfm_get_pmc_reset, PFM_CMD_ARG_RW, PFM_CMD_ARG_MANY, sizeof(pfarg_reg_t)},
/* 16 */{ NULL, 0, 0, 0}, /* not used */
/* 17 */{ NULL, 0, 0, 0}, /* not used */
/* 18 */{ NULL, 0, 0, 0}, /* not used */
/* 19 */{ NULL, 0, 0, 0}, /* not used */
/* 20 */{ NULL, 0, 0, 0}, /* not used */
/* 21 */{ NULL, 0, 0, 0}, /* not used */
/* 22 */{ NULL, 0, 0, 0}, /* not used */
/* 23 */{ NULL, 0, 0, 0}, /* not used */
/* 24 */{ NULL, 0, 0, 0}, /* not used */
/* 25 */{ NULL, 0, 0, 0}, /* not used */
/* 26 */{ NULL, 0, 0, 0}, /* not used */
/* 27 */{ NULL, 0, 0, 0}, /* not used */
/* 28 */{ NULL, 0, 0, 0}, /* not used */
/* 29 */{ NULL, 0, 0, 0}, /* not used */
/* 30 */{ NULL, 0, 0, 0}, /* not used */
/* 31 */{ NULL, 0, 0, 0}, /* not used */
#ifdef PFM_PMU_USES_DBR
/* 32 */{ pfm_write_ibrs, PFM_CMD_PID|PFM_CMD_CTX|PFM_CMD_ARG_RW, PFM_CMD_ARG_MANY, sizeof(pfarg_dbreg_t)},
/* 33 */{ pfm_write_dbrs, PFM_CMD_PID|PFM_CMD_CTX|PFM_CMD_ARG_RW, PFM_CMD_ARG_MANY, sizeof(pfarg_dbreg_t)}
#endif
};
#define PFM_CMD_COUNT (sizeof(pfm_cmd_tab)/sizeof(pfm_cmd_desc_t))
static int
check_task_state(struct task_struct *task)
{
int ret = 0;
#ifdef CONFIG_SMP
/* We must wait until the state has been completely
* saved. There can be situations where the reader arrives before
* after the task is marked as STOPPED but before pfm_save_regs()
* is completed.
*/
if (task->state != TASK_ZOMBIE && task->state != TASK_STOPPED) return -EBUSY;
DBprintk(("before wait_task_inactive [%d] state %ld\n", task->pid, task->state));
wait_task_inactive(task);
DBprintk(("after wait_task_inactive [%d] state %ld\n", task->pid, task->state));
#else
if (task->state != TASK_ZOMBIE && task->state != TASK_STOPPED) {
DBprintk(("warning [%d] not in stable state %ld\n", task->pid, task->state));
ret = -EBUSY;
}
#endif
return ret;
}
asmlinkage long
sys_perfmonctl (pid_t pid, int cmd, void *arg, int count, long arg5, long arg6, long arg7,
long arg8, long stack)
{
struct pt_regs *regs = (struct pt_regs *)&stack;
struct task_struct *task = current;
pfm_context_t *ctx;
size_t sz;
int ret, narg;
/*
* reject any call if perfmon was disabled at initialization time
*/
if (PFM_IS_DISABLED()) return -ENOSYS;
DBprintk(("cmd=%d idx=%d valid=%d narg=0x%x\n", cmd, PFM_CMD_IDX(cmd),
PFM_CMD_IS_VALID(cmd), PFM_CMD_NARG(cmd)));
if (PFM_CMD_IS_VALID(cmd) == 0) return -EINVAL;
/* ingore arguments when command has none */
narg = PFM_CMD_NARG(cmd);
if ((narg == PFM_CMD_ARG_MANY && count == 0) || (narg > 0 && narg != count)) return -EINVAL;
sz = PFM_CMD_ARG_SIZE(cmd);
if (PFM_CMD_READ_ARG(cmd) && !access_ok(VERIFY_READ, arg, sz*count)) return -EFAULT;
if (PFM_CMD_RW_ARG(cmd) && !access_ok(VERIFY_WRITE, arg, sz*count)) return -EFAULT;
if (PFM_CMD_USE_PID(cmd)) {
/*
* XXX: may need to fine tune this one
*/
if (pid < 2) return -EPERM;
if (pid != current->pid) {
ret = -ESRCH;
read_lock(&tasklist_lock);
task = find_task_by_pid(pid);
if (task) get_task_struct(task);
read_unlock(&tasklist_lock);
if (!task) goto abort_call;
ret = -EPERM;
if (pfm_bad_permissions(task)) goto abort_call;
if (PFM_CMD_CHK(cmd)) {
ret = check_task_state(task);
if (ret != 0) goto abort_call;
}
}
}
ctx = task->thread.pfm_context;
if (PFM_CMD_USE_CTX(cmd)) {
ret = -EINVAL;
if (ctx == NULL) {
DBprintk(("no context for task %d\n", task->pid));
goto abort_call;
}
ret = -EPERM;
/*
* we only grant access to the context if:
* - the caller is the creator of the context (ctx_owner)
* OR - the context is attached to the caller AND The context IS NOT
* in protected mode
*/
if (ctx->ctx_owner != current && (ctx->ctx_fl_protected || task != current)) {
DBprintk(("context protected, no access for [%d]\n", task->pid));
goto abort_call;
}
}
ret = (*pfm_cmd_tab[PFM_CMD_IDX(cmd)].cmd_func)(task, ctx, arg, count, regs);
abort_call:
if (task && task != current) put_task_struct(task);
return ret;
}
/*
* send SIGPROF to register task, must be invoked when it
* is safe to send a signal, e.g., not holding any runqueue
* related locks.
*/
static int
pfm_notify_user(pfm_context_t *ctx)
{
struct siginfo si;
int ret;
if (ctx->ctx_notify_task == NULL) {
DBprintk(("[%d] no notifier\n", current->pid));
return -EINVAL;
}
si.si_errno = 0;
si.si_addr = NULL;
si.si_pid = current->pid; /* who is sending */
si.si_signo = SIGPROF;
si.si_code = PROF_OVFL;
si.si_pfm_ovfl[0] = ctx->ctx_ovfl_regs[0];
/*
* when the target of the signal is not ourself, we have to be more
* careful. The notify_task may being cleared by the target task itself
* in release_thread(). We must ensure mutual exclusion here such that
* the signal is delivered (even to a dying task) safely.
*/
if (ctx->ctx_notify_task != current) {
/*
* grab the notification lock for this task
* This guarantees that the sequence: test + send_signal
* is atomic with regards to the ctx_notify_task field.
*
* We need a spinlock and not just an atomic variable for this.
*
*/
spin_lock(&ctx->ctx_lock);
/*
* now notify_task cannot be modified until we're done
* if NULL, they it got modified while we were in the handler
*/
if (ctx->ctx_notify_task == NULL) {
spin_unlock(&ctx->ctx_lock);
/*
* If we've lost the notified task, then we will run
* to completion wbut keep the PMU frozen. Results
* will be incorrect anyway. We do not kill task
* to leave it possible to attach perfmon context
* to already running task.
*/
printk("perfmon: pfm_notify_user() lost notify_task\n");
DBprintk_ovfl(("notification task has disappeared !\n"));
/* we cannot afford to block now */
ctx->ctx_fl_block = 0;
return -EINVAL;
}
/*
* required by send_sig_info() to make sure the target
* task does not disappear on us.
*/
read_lock(&tasklist_lock);
}
/*
* in this case, we don't stop the task, we let it go on. It will
* necessarily go to the signal handler (if any) when it goes back to
* user mode.
*/
DBprintk_ovfl(("[%d] sending notification to [%d]\n",
current->pid, ctx->ctx_notify_task->pid));
/*
* this call is safe in an interrupt handler, so does read_lock() on tasklist_lock
*/
ret = send_sig_info(SIGPROF, &si, ctx->ctx_notify_task);
if (ret) {
printk("perfmon: send_sig_info(process %d, SIGPROF)=%d\n",
ctx->ctx_notify_task->pid, ret);
}
/*
* now undo the protections in order
*/
if (ctx->ctx_notify_task != current) {
read_unlock(&tasklist_lock);
spin_unlock(&ctx->ctx_lock);
}
return ret;
}
void
pfm_ovfl_block_reset(void)
{
struct thread_struct *th = ¤t->thread;
pfm_context_t *ctx = current->thread.pfm_context;
unsigned int reason;
int ret;
/*
* clear the flag, to make sure we won't get here
* again
*/
th->pfm_ovfl_block_reset = 0;
/*
* do some sanity checks first
*/
if (!ctx) {
printk(KERN_DEBUG "perfmon: [%d] has no PFM context\n", current->pid);
return;
}
/*
* extract reason for being here and clear
*/
reason = ctx->ctx_fl_trap_reason;
ctx->ctx_fl_trap_reason = PFM_TRAP_REASON_NONE;
DBprintk(("[%d] reason=%d\n", current->pid, reason));
/*
* just here for a reset (non-blocking context only)
*/
if (reason == PFM_TRAP_REASON_RESET) goto non_blocking;
/*
* first notify user. This can fail if notify_task has disappeared.
*/
if (reason == PFM_TRAP_REASON_SIG || reason == PFM_TRAP_REASON_BLOCKSIG) {
ret = pfm_notify_user(ctx);
if (ret) return;
}
/*
* came here just to signal (non-blocking)
*/
if (reason == PFM_TRAP_REASON_SIG) return;
DBprintk(("[%d] before sleeping\n", current->pid));
/*
* may go through without blocking on SMP systems
* if restart has been received already by the time we call down()
*/
ret = down_interruptible(&ctx->ctx_restart_sem);
DBprintk(("[%d] after sleeping ret=%d\n", current->pid, ret));
/*
* in case of interruption of down() we don't restart anything
*/
if (ret >= 0) {
non_blocking:
/* we reactivate on context switch */
ctx->ctx_fl_frozen = 0;
/*
* the ovfl_sem is cleared by the restart task and this is safe because we always
* use the local reference
*/
pfm_reset_regs(ctx, ctx->ctx_ovfl_regs, PFM_PMD_LONG_RESET);
ctx->ctx_ovfl_regs[0] = 0UL;
/*
* Unlock sampling buffer and reset index atomically
* XXX: not really needed when blocking
*/
if (CTX_HAS_SMPL(ctx)) {
ctx->ctx_psb->psb_hdr->hdr_count = 0;
ctx->ctx_psb->psb_index = 0;
}
pfm_unfreeze_pmu();
/* state restored, can go back to work (user mode) */
}
}
/*
* This function will record an entry in the sampling if it is not full already.
* Return:
* 0 : buffer is not full (did not BECOME full: still space or was already full)
* 1 : buffer is full (recorded the last entry)
*/
static int
pfm_record_sample(struct task_struct *task, pfm_context_t *ctx, unsigned long ovfl_mask, struct pt_regs *regs)
{
pfm_smpl_buffer_desc_t *psb = ctx->ctx_psb;
unsigned long *e, m, idx;
perfmon_smpl_entry_t *h;
int j;
idx = ia64_fetch_and_add(1, &psb->psb_index);
DBprintk_ovfl(("recording index=%ld entries=%ld\n", idx-1, psb->psb_entries));
/*
* XXX: there is a small chance that we could run out on index before resetting
* but index is unsigned long, so it will take some time.....
* We use > instead of == because fetch_and_add() is off by one (see below)
*
* This case can happen in non-blocking mode or with multiple processes.
* For non-blocking, we need to reload and continue.
*/
if (idx > psb->psb_entries) return 0;
/* first entry is really entry 0, not 1 caused by fetch_and_add */
idx--;
h = (perfmon_smpl_entry_t *)(((char *)psb->psb_addr) + idx*(psb->psb_entry_size));
/*
* initialize entry header
*/
h->pid = current->pid;
h->cpu = get_cpu();
h->last_reset_value = ovfl_mask ? ctx->ctx_soft_pmds[ffz(~ovfl_mask)].lval : 0UL;
h->ip = regs ? regs->cr_iip | ((regs->cr_ipsr >> 41) & 0x3): 0x0UL;
h->regs = ovfl_mask; /* which registers overflowed */
/* guaranteed to monotonically increase on each cpu */
h->stamp = pfm_get_stamp();
/* position for first pmd */
e = (unsigned long *)(h+1);
/*
* selectively store PMDs in increasing index number
*/
m = ctx->ctx_smpl_regs[0];
for (j=0; m; m >>=1, j++) {
if ((m & 0x1) == 0) continue;
if (PMD_IS_COUNTING(j)) {
*e = pfm_read_soft_counter(ctx, j);
} else {
*e = ia64_get_pmd(j); /* slow */
}
DBprintk_ovfl(("e=%p pmd%d =0x%lx\n", (void *)e, j, *e));
e++;
}
pfm_stats[h->cpu].pfm_recorded_samples_count++;
/*
* make the new entry visible to user, needs to be atomic
*/
ia64_fetch_and_add(1, &psb->psb_hdr->hdr_count);
DBprintk_ovfl(("index=%ld entries=%ld hdr_count=%ld\n",
idx, psb->psb_entries, psb->psb_hdr->hdr_count));
/*
* sampling buffer full ?
*/
if (idx == (psb->psb_entries-1)) {
DBprintk_ovfl(("sampling buffer full\n"));
/*
* XXX: must reset buffer in blocking mode and lost notified
*/
pfm_stats[h->cpu].pfm_full_smpl_buffer_count++;
put_cpu();
return 1;
}
put_cpu();
return 0;
}
/*
* main overflow processing routine.
* it can be called from the interrupt path or explicitely during the context switch code
* Return:
* new value of pmc[0]. if 0x0 then unfreeze, else keep frozen
*/
static unsigned long
pfm_overflow_handler(struct task_struct *task, pfm_context_t *ctx, u64 pmc0, struct pt_regs *regs)
{
unsigned long mask;
struct thread_struct *t;
unsigned long old_val;
unsigned long ovfl_notify = 0UL, ovfl_pmds = 0UL;
int i;
int ret = 1;
/*
* It is never safe to access the task for which the overflow interrupt is destinated
* using the current variable as the interrupt may occur in the middle of a context switch
* where current does not hold the task that is running yet.
*
* For monitoring, however, we do need to get access to the task which caused the overflow
* to account for overflow on the counters.
*
* We accomplish this by maintaining a current owner of the PMU per CPU. During context
* switch the ownership is changed in a way such that the reflected owner is always the
* valid one, i.e. the one that caused the interrupt.
*/
preempt_disable();
t = &task->thread;
/*
* XXX: debug test
* Don't think this could happen given upfront tests
*/
if ((t->flags & IA64_THREAD_PM_VALID) == 0 && ctx->ctx_fl_system == 0) {
printk(KERN_DEBUG "perfmon: Spurious overflow interrupt: process %d not "
"using perfmon\n", task->pid);
preempt_enable_no_resched();
return 0x1;
}
/*
* sanity test. Should never happen
*/
if ((pmc0 & 0x1) == 0) {
printk(KERN_DEBUG "perfmon: pid %d pmc0=0x%lx assumption error for freeze bit\n",
task->pid, pmc0);
preempt_enable_no_resched();
return 0x0;
}
mask = pmc0 >> PMU_FIRST_COUNTER;
DBprintk_ovfl(("pmc0=0x%lx pid=%d iip=0x%lx, %s"
" mode used_pmds=0x%lx used_pmcs=0x%lx reload_pmcs=0x%lx\n",
pmc0, task->pid, (regs ? regs->cr_iip : 0),
CTX_OVFL_NOBLOCK(ctx) ? "nonblocking" : "blocking",
ctx->ctx_used_pmds[0],
ctx->ctx_used_pmcs[0],
ctx->ctx_reload_pmcs[0]));
/*
* First we update the virtual counters
*/
for (i = PMU_FIRST_COUNTER; mask ; i++, mask >>= 1) {
/* skip pmd which did not overflow */
if ((mask & 0x1) == 0) continue;
DBprintk_ovfl(("pmd[%d] overflowed hw_pmd=0x%lx soft_pmd=0x%lx\n",
i, ia64_get_pmd(i), ctx->ctx_soft_pmds[i].val));
/*
* Note that the pmd is not necessarily 0 at this point as qualified events
* may have happened before the PMU was frozen. The residual count is not
* taken into consideration here but will be with any read of the pmd via
* pfm_read_pmds().
*/
old_val = ctx->ctx_soft_pmds[i].val;
ctx->ctx_soft_pmds[i].val += 1 + pmu_conf.ovfl_val;
/*
* check for overflow condition
*/
if (old_val > ctx->ctx_soft_pmds[i].val) {
ovfl_pmds |= 1UL << i;
if (PMC_OVFL_NOTIFY(ctx, i)) {
ovfl_notify |= 1UL << i;
}
}
DBprintk_ovfl(("soft_pmd[%d].val=0x%lx old_val=0x%lx pmd=0x%lx ovfl_pmds=0x%lx ovfl_notify=0x%lx\n",
i, ctx->ctx_soft_pmds[i].val, old_val,
ia64_get_pmd(i) & pmu_conf.ovfl_val, ovfl_pmds, ovfl_notify));
}
/*
* check for sampling buffer
*
* if present, record sample. We propagate notification ONLY when buffer
* becomes full.
*/
if(CTX_HAS_SMPL(ctx)) {
ret = pfm_record_sample(task, ctx, ovfl_pmds, regs);
if (ret == 1) {
/*
* Sampling buffer became full
* If no notication was requested, then we reset buffer index
* and reset registers (done below) and resume.
* If notification requested, then defer reset until pfm_restart()
*/
if (ovfl_notify == 0UL) {
ctx->ctx_psb->psb_hdr->hdr_count = 0UL;
ctx->ctx_psb->psb_index = 0UL;
}
} else {
/*
* sample recorded in buffer, no need to notify user
*/
ovfl_notify = 0UL;
}
}
/*
* No overflow requiring a user level notification
*/
if (ovfl_notify == 0UL) {
if (ovfl_pmds)
pfm_reset_regs(ctx, &ovfl_pmds, PFM_PMD_SHORT_RESET);
preempt_enable_no_resched();
return 0x0UL;
}
/*
* keep track of what to reset when unblocking
*/
ctx->ctx_ovfl_regs[0] = ovfl_pmds;
DBprintk_ovfl(("block=%d notify [%d] current [%d]\n",
ctx->ctx_fl_block,
ctx->ctx_notify_task ? ctx->ctx_notify_task->pid: -1,
current->pid ));
/*
* ctx_notify_task could already be NULL, checked in pfm_notify_user()
*/
if (CTX_OVFL_NOBLOCK(ctx) == 0 && ctx->ctx_notify_task != task) {
t->pfm_ovfl_block_reset = 1; /* will cause blocking */
ctx->ctx_fl_trap_reason = PFM_TRAP_REASON_BLOCKSIG;
} else {
t->pfm_ovfl_block_reset = 1; /* will cause blocking */
ctx->ctx_fl_trap_reason = PFM_TRAP_REASON_SIG;
}
/*
* keep the PMU frozen until either pfm_restart() or
* task completes (non-blocking or notify_task gone).
*/
ctx->ctx_fl_frozen = 1;
DBprintk_ovfl(("return pmc0=0x%x must_block=%ld reason=%d\n",
ctx->ctx_fl_frozen ? 0x1 : 0x0,
t->pfm_ovfl_block_reset,
ctx->ctx_fl_trap_reason));
preempt_enable_no_resched();
return 0x1UL;
}
static void
pfm_interrupt_handler(int irq, void *arg, struct pt_regs *regs)
{
u64 pmc0;
struct task_struct *task;
pfm_context_t *ctx;
pfm_stats[get_cpu()].pfm_ovfl_intr_count++;
/*
* if an alternate handler is registered, just bypass the default one
*/
if (pfm_alternate_intr_handler) {
(*pfm_alternate_intr_handler->handler)(irq, arg, regs);
put_cpu();
return;
}
/*
* srlz.d done before arriving here
*
* This is slow
*/
pmc0 = ia64_get_pmc(0);
/*
* if we have some pending bits set
* assumes : if any PM[0].bit[63-1] is set, then PMC[0].fr = 1
*/
if ((pmc0 & ~0x1UL)!=0UL && (task=PMU_OWNER())!= NULL) {
/*
* we assume that pmc0.fr is always set here
*/
ctx = task->thread.pfm_context;
/* sanity check */
if (!ctx) {
printk(KERN_DEBUG "perfmon: Spurious overflow interrupt: process %d has "
"no PFM context\n", task->pid);
put_cpu();
return;
}
/*
* assume PMC[0].fr = 1 at this point
*/
pmc0 = pfm_overflow_handler(task, ctx, pmc0, regs);
/*
* we can only update pmc0 when the overflow
* is for the current context. In UP the current
* task may not be the one owning the PMU
*/
if (task == current) {
/*
* We always clear the overflow status bits and either unfreeze
* or keep the PMU frozen.
*/
ia64_set_pmc(0, pmc0);
ia64_srlz_d();
} else {
task->thread.pmc[0] = pmc0;
}
} else {
pfm_stats[smp_processor_id()].pfm_spurious_ovfl_intr_count++;
}
put_cpu_no_resched();
}
/* for debug only */
static int
pfm_proc_info(char *page)
{
char *p = page;
int i;
p += sprintf(p, "fastctxsw : %s\n", pfm_sysctl.fastctxsw > 0 ? "Yes": "No");
p += sprintf(p, "ovfl_mask : 0x%lx\n", pmu_conf.ovfl_val);
for(i=0; i < NR_CPUS; i++) {
if (cpu_is_online(i) == 0) continue;
p += sprintf(p, "CPU%-2d overflow intrs : %lu\n", i, pfm_stats[i].pfm_ovfl_intr_count);
p += sprintf(p, "CPU%-2d spurious intrs : %lu\n", i, pfm_stats[i].pfm_spurious_ovfl_intr_count);
p += sprintf(p, "CPU%-2d recorded samples : %lu\n", i, pfm_stats[i].pfm_recorded_samples_count);
p += sprintf(p, "CPU%-2d smpl buffer full : %lu\n", i, pfm_stats[i].pfm_full_smpl_buffer_count);
p += sprintf(p, "CPU%-2d syst_wide : %d\n", i, per_cpu(pfm_syst_info, i) & PFM_CPUINFO_SYST_WIDE ? 1 : 0);
p += sprintf(p, "CPU%-2d dcr_pp : %d\n", i, per_cpu(pfm_syst_info, i) & PFM_CPUINFO_DCR_PP ? 1 : 0);
p += sprintf(p, "CPU%-2d exclude idle : %d\n", i, per_cpu(pfm_syst_info, i) & PFM_CPUINFO_EXCL_IDLE ? 1 : 0);
p += sprintf(p, "CPU%-2d owner : %d\n", i, pmu_owners[i].owner ? pmu_owners[i].owner->pid: -1);
}
LOCK_PFS();
p += sprintf(p, "proc_sessions : %u\n"
"sys_sessions : %u\n"
"sys_use_dbregs : %u\n"
"ptrace_use_dbregs : %u\n",
pfm_sessions.pfs_task_sessions,
pfm_sessions.pfs_sys_sessions,
pfm_sessions.pfs_sys_use_dbregs,
pfm_sessions.pfs_ptrace_use_dbregs);
UNLOCK_PFS();
return p - page;
}
/* /proc interface, for debug only */
static int
perfmon_read_entry(char *page, char **start, off_t off, int count, int *eof, void *data)
{
int len = pfm_proc_info(page);
if (len <= off+count) *eof = 1;
*start = page + off;
len -= off;
if (len>count) len = count;
if (len<0) len = 0;
return len;
}
/*
* we come here as soon as PFM_CPUINFO_SYST_WIDE is set. This happens
* during pfm_enable() hence before pfm_start(). We cannot assume monitoring
* is active or inactive based on mode. We must rely on the value in
* cpu_data(i)->pfm_syst_info
*/
void
pfm_syst_wide_update_task(struct task_struct *task, unsigned long info, int is_ctxswin)
{
struct pt_regs *regs;
unsigned long dcr;
unsigned long dcr_pp;
preempt_disable();
dcr_pp = info & PFM_CPUINFO_DCR_PP ? 1 : 0;
/*
* pid 0 is guaranteed to be the idle task. There is one such task with pid 0
* on every CPU, so we can rely on the pid to identify the idle task.
*/
if ((info & PFM_CPUINFO_EXCL_IDLE) == 0 || task->pid) {
regs = (struct pt_regs *)((unsigned long) task + IA64_STK_OFFSET);
regs--;
ia64_psr(regs)->pp = is_ctxswin ? dcr_pp : 0;
preempt_enable();
return;
}
/*
* if monitoring has started
*/
if (dcr_pp) {
dcr = ia64_get_dcr();
/*
* context switching in?
*/
if (is_ctxswin) {
/* mask monitoring for the idle task */
ia64_set_dcr(dcr & ~IA64_DCR_PP);
pfm_clear_psr_pp();
ia64_srlz_i();
preempt_enable();
return;
}
/*
* context switching out
* restore monitoring for next task
*
* Due to inlining this odd if-then-else construction generates
* better code.
*/
ia64_set_dcr(dcr |IA64_DCR_PP);
pfm_set_psr_pp();
ia64_srlz_i();
}
preempt_enable();
}
void
pfm_save_regs (struct task_struct *task)
{
pfm_context_t *ctx;
unsigned long mask;
u64 psr;
int i;
preempt_disable();
ctx = task->thread.pfm_context;
/*
* save current PSR: needed because we modify it
*/
psr = pfm_get_psr();
/*
* stop monitoring:
* This is the last instruction which can generate an overflow
*
* We do not need to set psr.sp because, it is irrelevant in kernel.
* It will be restored from ipsr when going back to user level
*/
pfm_clear_psr_up();
ia64_srlz_i();
ctx->ctx_saved_psr = psr;
#ifdef CONFIG_SMP
/*
* We do not use a lazy scheme in SMP because
* of the new scheduler which masks interrupts
* during low-level context switch. So we save
* all the PMD register we use and restore on
* ctxsw in.
*
* release ownership of this PMU.
* must be done before we save the registers.
*/
SET_PMU_OWNER(NULL);
/*
* save PMDs
*/
ia64_srlz_d();
mask = ctx->ctx_used_pmds[0];
for (i=0; mask; i++, mask>>=1) {
if (mask & 0x1) task->thread.pmd[i] =ia64_get_pmd(i);
}
/*
* save pmc0
*/
task->thread.pmc[0] = ia64_get_pmc(0);
/*
* force a full reload
*/
atomic_set(&ctx->ctx_last_cpu, -1);
#endif
preempt_enable();
}
static void
pfm_lazy_save_regs (struct task_struct *task)
{
pfm_context_t *ctx;
struct thread_struct *t;
unsigned long mask;
int i;
preempt_disable();
DBprintk(("on [%d] by [%d]\n", task->pid, current->pid));
t = &task->thread;
ctx = task->thread.pfm_context;
/*
* do not own the PMU
*/
SET_PMU_OWNER(NULL);
ia64_srlz_d();
/*
* XXX needs further optimization.
* Also must take holes into account
*/
mask = ctx->ctx_used_pmds[0];
for (i=0; mask; i++, mask>>=1) {
if (mask & 0x1) t->pmd[i] =ia64_get_pmd(i);
}
/* save pmc0 */
t->pmc[0] = ia64_get_pmc(0);
/* not owned by this CPU */
atomic_set(&ctx->ctx_last_cpu, -1);
preempt_enable();
}
void
pfm_load_regs (struct task_struct *task)
{
struct thread_struct *t;
pfm_context_t *ctx;
struct task_struct *owner;
unsigned long mask;
u64 psr;
int i;
preempt_disable();
owner = PMU_OWNER();
ctx = task->thread.pfm_context;
t = &task->thread;
if (ctx == NULL) {
preempt_enable();
printk("perfmon: pfm_load_regs: null ctx for [%d]\n", task->pid);
return;
}
/*
* we restore ALL the debug registers to avoid picking up
* stale state.
*
* This must be done even when the task is still the owner
* as the registers may have been modified via ptrace()
* (not perfmon) by the previous task.
*
* XXX: dealing with this in a lazy fashion requires modifications
* to the way the the debug registers are managed. This is will done
* in the next version of perfmon.
*/
if (ctx->ctx_fl_using_dbreg) {
for (i=0; i < pmu_conf.num_ibrs; i++) {
ia64_set_ibr(i, t->ibr[i]);
}
ia64_srlz_i();
for (i=0; i < pmu_conf.num_dbrs; i++) {
ia64_set_dbr(i, t->dbr[i]);
}
ia64_srlz_d();
}
/*
* if we were the last user, then nothing to do except restore psr
* this path cannot be used in SMP
*/
if (owner == task) {
if (atomic_read(&ctx->ctx_last_cpu) != smp_processor_id())
DBprintk(("invalid last_cpu=%d for [%d]\n",
atomic_read(&ctx->ctx_last_cpu), task->pid));
psr = ctx->ctx_saved_psr;
pfm_set_psr_l(psr);
preempt_enable();
return;
}
/*
* someone else is still using the PMU, first push it out and
* then we'll be able to install our stuff !
*
* not possible in SMP
*/
if (owner) pfm_lazy_save_regs(owner);
/*
* To avoid leaking information to the user level when psr.sp=0,
* we must reload ALL implemented pmds (even the ones we don't use).
* In the kernel we only allow PFM_READ_PMDS on registers which
* we initialized or requested (sampling) so there is no risk there.
*
* As an optimization, we will only reload the PMD that we use when
* the context is in protected mode, i.e. psr.sp=1 because then there
* is no leak possible.
*/
mask = pfm_sysctl.fastctxsw || ctx->ctx_fl_protected ? ctx->ctx_used_pmds[0] : ctx->ctx_reload_pmds[0];
for (i=0; mask; i++, mask>>=1) {
if (mask & 0x1) ia64_set_pmd(i, t->pmd[i] & pmu_conf.ovfl_val);
}
/*
* PMC0 is never set in the mask because it is always restored
* separately.
*
* ALL PMCs are systematically reloaded, unused registers
* get their default (PAL reset) values to avoid picking up
* stale configuration.
*/
mask = ctx->ctx_reload_pmcs[0];
for (i=0; mask; i++, mask>>=1) {
if (mask & 0x1) ia64_set_pmc(i, t->pmc[i]);
}
/*
* manually invoke core interrupt handler
* if the task had a pending overflow when it was ctxsw out.
* Side effect on ctx_fl_frozen is possible.
*/
if (t->pmc[0] & ~0x1) {
t->pmc[0] = pfm_overflow_handler(task, ctx, t->pmc[0], NULL);
}
/*
* unfreeze PMU if possible
*/
if (ctx->ctx_fl_frozen == 0) pfm_unfreeze_pmu();
atomic_set(&ctx->ctx_last_cpu, smp_processor_id());
SET_PMU_OWNER(task);
/*
* restore the psr we changed in pfm_save_regs()
*/
psr = ctx->ctx_saved_psr;
preempt_enable();
pfm_set_psr_l(psr);
}
/*
* XXX: make this routine able to work with non current context
*/
static void
pfm_reset_pmu(struct task_struct *task)
{
struct thread_struct *t = &task->thread;
pfm_context_t *ctx = t->pfm_context;
int i;
if (task != current) {
printk("perfmon: invalid task in pfm_reset_pmu()\n");
return;
}
preempt_disable();
/* Let's make sure the PMU is frozen */
pfm_freeze_pmu();
/*
* install reset values for PMC. We skip PMC0 (done above)
* XX: good up to 64 PMCS
*/
for (i=1; (pmu_conf.pmc_desc[i].type & PFM_REG_END) == 0; i++) {
if ((pmu_conf.pmc_desc[i].type & PFM_REG_IMPL) == 0) continue;
ia64_set_pmc(i, PMC_DFL_VAL(i));
/*
* When restoring context, we must restore ALL pmcs, even the ones
* that the task does not use to avoid leaks and possibly corruption
* of the sesion because of configuration conflicts. So here, we
* initialize the entire set used in the context switch restore routine.
*/
t->pmc[i] = PMC_DFL_VAL(i);
DBprintk(("pmc[%d]=0x%lx\n", i, t->pmc[i]));
}
/*
* clear reset values for PMD.
* XXX: good up to 64 PMDS.
*/
for (i=0; (pmu_conf.pmd_desc[i].type & PFM_REG_END) == 0; i++) {
if ((pmu_conf.pmd_desc[i].type & PFM_REG_IMPL) == 0) continue;
ia64_set_pmd(i, 0UL);
t->pmd[i] = 0UL;
}
/*
* On context switched restore, we must restore ALL pmc and ALL pmd even
* when they are not actively used by the task. In UP, the incoming process
* may otherwise pick up left over PMC, PMD state from the previous process.
* As opposed to PMD, stale PMC can cause harm to the incoming
* process because they may change what is being measured.
* Therefore, we must systematically reinstall the entire
* PMC state. In SMP, the same thing is possible on the
* same CPU but also on between 2 CPUs.
*
* The problem with PMD is information leaking especially
* to user level when psr.sp=0
*
* There is unfortunately no easy way to avoid this problem
* on either UP or SMP. This definitively slows down the
* pfm_load_regs() function.
*/
/*
* We must include all the PMC in this mask to make sure we don't
* see any side effect of a stale state, such as opcode matching
* or range restrictions, for instance.
*
* We never directly restore PMC0 so we do not include it in the mask.
*/
ctx->ctx_reload_pmcs[0] = pmu_conf.impl_pmcs[0] & ~0x1;
/*
* We must include all the PMD in this mask to avoid picking
* up stale value and leak information, especially directly
* at the user level when psr.sp=0
*/
ctx->ctx_reload_pmds[0] = pmu_conf.impl_pmds[0];
/*
* Keep track of the pmds we want to sample
* XXX: may be we don't need to save/restore the DEAR/IEAR pmds
* but we do need the BTB for sure. This is because of a hardware
* buffer of 1 only for non-BTB pmds.
*
* We ignore the unimplemented pmds specified by the user
*/
ctx->ctx_used_pmds[0] = ctx->ctx_smpl_regs[0];
ctx->ctx_used_pmcs[0] = 1; /* always save/restore PMC[0] */
/*
* useful in case of re-enable after disable
*/
ctx->ctx_used_ibrs[0] = 0UL;
ctx->ctx_used_dbrs[0] = 0UL;
ia64_srlz_d();
preempt_enable();
}
/*
* This function is called when a thread exits (from exit_thread()).
* This is a simplified pfm_save_regs() that simply flushes the current
* register state into the save area taking into account any pending
* overflow. This time no notification is sent because the task is dying
* anyway. The inline processing of overflows avoids loosing some counts.
* The PMU is frozen on exit from this call and is to never be reenabled
* again for this task.
*
*/
void
pfm_flush_regs (struct task_struct *task)
{
pfm_context_t *ctx;
u64 pmc0;
unsigned long mask2, val;
int i;
ctx = task->thread.pfm_context;
if (ctx == NULL) return;
/*
* that's it if context already disabled
*/
if (ctx->ctx_flags.state == PFM_CTX_DISABLED) return;
preempt_disable();
/*
* stop monitoring:
* This is the only way to stop monitoring without destroying overflow
* information in PMC[0].
* This is the last instruction which can cause overflow when monitoring
* in kernel.
* By now, we could still have an overflow interrupt in-flight.
*/
if (ctx->ctx_fl_system) {
/* disable dcr pp */
ia64_set_dcr(ia64_get_dcr() & ~IA64_DCR_PP);
/* stop monitoring */
pfm_clear_psr_pp();
ia64_srlz_i();
PFM_CPUINFO_CLEAR(PFM_CPUINFO_SYST_WIDE);
PFM_CPUINFO_CLEAR(PFM_CPUINFO_DCR_PP);
PFM_CPUINFO_CLEAR(PFM_CPUINFO_EXCL_IDLE);
} else {
/* stop monitoring */
pfm_clear_psr_up();
ia64_srlz_i();
/* no more save/restore on ctxsw */
current->thread.flags &= ~IA64_THREAD_PM_VALID;
}
/*
* Mark the PMU as not owned
* This will cause the interrupt handler to do nothing in case an overflow
* interrupt was in-flight
* This also guarantees that pmc0 will contain the final state
* It virtually gives us full control on overflow processing from that point
* on.
* It must be an atomic operation.
*/
SET_PMU_OWNER(NULL);
/*
* read current overflow status:
*
* we are guaranteed to read the final stable state
*/
ia64_srlz_d();
pmc0 = ia64_get_pmc(0); /* slow */
/*
* freeze PMU:
*
* This destroys the overflow information. This is required to make sure
* next process does not start with monitoring on if not requested
*/
pfm_freeze_pmu();
/*
* We don't need to restore psr, because we are on our way out
*/
/*
* This loop flushes the PMD into the PFM context.
* It also processes overflow inline.
*
* IMPORTANT: No notification is sent at this point as the process is dying.
* The implicit notification will come from a SIGCHILD or a return from a
* waitpid().
*
*/
if (atomic_read(&ctx->ctx_last_cpu) != smp_processor_id())
printk(KERN_DEBUG "perfmon: [%d] last_cpu=%d\n",
task->pid, atomic_read(&ctx->ctx_last_cpu));
/*
* we save all the used pmds
* we take care of overflows for pmds used as counters
*/
mask2 = ctx->ctx_used_pmds[0];
for (i = 0; mask2; i++, mask2>>=1) {
/* skip non used pmds */
if ((mask2 & 0x1) == 0) continue;
val = ia64_get_pmd(i);
if (PMD_IS_COUNTING(i)) {
DBprintk(("[%d] pmd[%d] soft_pmd=0x%lx hw_pmd=0x%lx\n",
task->pid,
i,
ctx->ctx_soft_pmds[i].val,
val & pmu_conf.ovfl_val));
/* collect latest results */
ctx->ctx_soft_pmds[i].val += val & pmu_conf.ovfl_val;
/*
* now everything is in ctx_soft_pmds[] and we need
* to clear the saved context from save_regs() such that
* pfm_read_pmds() gets the correct value
*/
task->thread.pmd[i] = 0;
/*
* take care of overflow inline
*/
if (pmc0 & (1UL << i)) {
ctx->ctx_soft_pmds[i].val += 1 + pmu_conf.ovfl_val;
DBprintk(("[%d] pmd[%d] overflowed soft_pmd=0x%lx\n",
task->pid, i, ctx->ctx_soft_pmds[i].val));
}
} else {
DBprintk(("[%d] pmd[%d] hw_pmd=0x%lx\n", task->pid, i, val));
/*
* not a counter, just save value as is
*/
task->thread.pmd[i] = val;
}
}
/*
* indicates that context has been saved
*/
atomic_set(&ctx->ctx_last_cpu, -1);
preempt_enable();
}
/*
* task is the newly created task, pt_regs for new child
*/
int
pfm_inherit(struct task_struct *task, struct pt_regs *regs)
{
pfm_context_t *ctx;
pfm_context_t *nctx;
struct thread_struct *thread;
unsigned long m;
int i;
/*
* the new task was copied from parent and therefore points
* to the parent's context at this point
*/
ctx = task->thread.pfm_context;
thread = &task->thread;
preempt_disable();
/*
* make sure child cannot mess up the monitoring session
*/
ia64_psr(regs)->sp = 1;
DBprintk(("enabling psr.sp for [%d]\n", task->pid));
/*
* if there was a virtual mapping for the sampling buffer
* the mapping is NOT inherited across fork() (see VM_DONTCOPY),
* so we don't have to explicitely remove it here.
*
*
* Part of the clearing of fields is also done in
* copy_thread() because the fiels are outside the
* pfm_context structure and can affect tasks not
* using perfmon.
*/
/* clear pending notification */
task->thread.pfm_ovfl_block_reset = 0;
/*
* clear cpu pinning restriction for child
*/
if (ctx->ctx_fl_system) {
set_cpus_allowed(task, ctx->ctx_saved_cpus_allowed);
DBprintk(("setting cpus_allowed for [%d] to 0x%lx from 0x%lx\n",
task->pid,
ctx->ctx_saved_cpus_allowed,
current->cpus_allowed));
}
/*
* takes care of easiest case first
*/
if (CTX_INHERIT_MODE(ctx) == PFM_FL_INHERIT_NONE) {
DBprintk(("removing PFM context for [%d]\n", task->pid));
task->thread.pfm_context = NULL;
/*
* we must clear psr.up because the new child does
* not have a context and the PM_VALID flag is cleared
* in copy_thread().
*
* we do not clear psr.pp because it is always
* controlled by the system wide logic and we should
* never be here when system wide is running anyway
*/
ia64_psr(regs)->up = 0;
preempt_enable();
/* copy_thread() clears IA64_THREAD_PM_VALID */
return 0;
}
nctx = pfm_context_alloc();
if (nctx == NULL) return -ENOMEM;
/* copy content */
*nctx = *ctx;
if (CTX_INHERIT_MODE(ctx) == PFM_FL_INHERIT_ONCE) {
nctx->ctx_fl_inherit = PFM_FL_INHERIT_NONE;
DBprintk(("downgrading to INHERIT_NONE for [%d]\n", task->pid));
}
/*
* task is not yet visible in the tasklist, so we do
* not need to lock the newly created context.
* However, we must grab the tasklist_lock to ensure
* that the ctx_owner or ctx_notify_task do not disappear
* while we increment their check counters.
*/
read_lock(&tasklist_lock);
if (nctx->ctx_notify_task)
atomic_inc(&nctx->ctx_notify_task->thread.pfm_notifiers_check);
if (nctx->ctx_owner)
atomic_inc(&nctx->ctx_owner->thread.pfm_owners_check);
read_unlock(&tasklist_lock);
LOCK_PFS();
pfm_sessions.pfs_task_sessions++;
UNLOCK_PFS();
/* initialize counters in new context */
m = nctx->ctx_used_pmds[0] >> PMU_FIRST_COUNTER;
for(i = PMU_FIRST_COUNTER ; m ; m>>=1, i++) {
if ((m & 0x1) && pmu_conf.pmd_desc[i].type == PFM_REG_COUNTING) {
nctx->ctx_soft_pmds[i].val = nctx->ctx_soft_pmds[i].lval & ~pmu_conf.ovfl_val;
thread->pmd[i] = nctx->ctx_soft_pmds[i].lval & pmu_conf.ovfl_val;
} else {
thread->pmd[i] = 0UL; /* reset to initial state */
}
}
nctx->ctx_fl_frozen = 0;
nctx->ctx_ovfl_regs[0] = 0UL;
nctx->ctx_fl_trap_reason = PFM_TRAP_REASON_NONE;
atomic_set(&nctx->ctx_last_cpu, -1);
/*
* here nctx->ctx_psb == ctx->ctx_psb
*
* increment reference count to sampling
* buffer, if any. Note that this is independent
* from the virtual mapping. The latter is never
* inherited while the former will be if context
* is setup to something different from PFM_FL_INHERIT_NONE
*/
if (nctx->ctx_psb) {
LOCK_PSB(nctx->ctx_psb);
nctx->ctx_psb->psb_refcnt++;
DBprintk(("updated smpl @ %p refcnt=%lu psb_flags=0x%x\n",
ctx->ctx_psb->psb_hdr,
ctx->ctx_psb->psb_refcnt,
ctx->ctx_psb->psb_flags));
UNLOCK_PSB(nctx->ctx_psb);
/*
* remove any pointer to sampling buffer mapping
*/
nctx->ctx_smpl_vaddr = 0;
}
sema_init(&nctx->ctx_restart_sem, 0); /* reset this semaphore to locked */
/*
* propagate kernel psr in new context (used for first ctxsw in
*/
nctx->ctx_saved_psr = pfm_get_psr();
/*
* propagate kernel psr in new context (used for first ctxsw in
*/
nctx->ctx_saved_psr = pfm_get_psr();
/* link with new task */
thread->pfm_context = nctx;
DBprintk(("nctx=%p for process [%d]\n", (void *)nctx, task->pid));
/*
* the copy_thread routine automatically clears
* IA64_THREAD_PM_VALID, so we need to reenable it, if it was used by the caller
*/
if (current->thread.flags & IA64_THREAD_PM_VALID) {
DBprintk(("setting PM_VALID for [%d]\n", task->pid));
thread->flags |= IA64_THREAD_PM_VALID;
}
preempt_enable();
return 0;
}
/*
*
* We cannot touch any of the PMU registers at this point as we may
* not be running on the same CPU the task was last run on. Therefore
* it is assumed that the PMU has been stopped appropriately in
* pfm_flush_regs() called from exit_thread().
*
* The function is called in the context of the parent via a release_thread()
* and wait4(). The task is not in the tasklist anymore.
*/
void
pfm_context_exit(struct task_struct *task)
{
pfm_context_t *ctx = task->thread.pfm_context;
/*
* check sampling buffer
*/
preempt_disable();
if (ctx->ctx_psb) {
pfm_smpl_buffer_desc_t *psb = ctx->ctx_psb;
LOCK_PSB(psb);
DBprintk(("sampling buffer from [%d] @%p size %ld refcnt=%lu psb_flags=0x%x\n",
task->pid,
psb->psb_hdr, psb->psb_size, psb->psb_refcnt, psb->psb_flags));
/*
* in the case where we are the last user, we may be able to free
* the buffer
*/
psb->psb_refcnt--;
if (psb->psb_refcnt == 0) {
/*
* The flag is cleared in pfm_vm_close(). which gets
* called from do_exit() via exit_mm().
* By the time we come here, the task has no more mm context.
*
* We can only free the psb and buffer here after the vm area
* describing the buffer has been removed. This normally happens
* as part of do_exit() but the entire mm context is ONLY removed
* once its reference counts goes to zero. This is typically
* the case except for multi-threaded (several tasks) processes.
*
* See pfm_vm_close() and pfm_cleanup_smpl_buf() for more details.
*/
if ((psb->psb_flags & PSB_HAS_VMA) == 0) {
DBprintk(("cleaning sampling buffer from [%d] @%p size %ld\n",
task->pid,
psb->psb_hdr, psb->psb_size));
/*
* free the buffer and psb
*/
pfm_rvfree(psb->psb_hdr, psb->psb_size);
kfree(psb);
psb = NULL;
}
}
/* psb may have been deleted */
if (psb) UNLOCK_PSB(psb);
}
DBprintk(("cleaning [%d] pfm_context @%p notify_task=%p check=%d mm=%p\n",
task->pid, ctx,
ctx->ctx_notify_task,
atomic_read(&task->thread.pfm_notifiers_check), task->mm));
/*
* To avoid getting the notified task or owner task scan the entire process
* list when they exit, we decrement notifiers_check and owners_check respectively.
*
* Of course, there is race condition between decreasing the value and the
* task exiting. The danger comes from the fact that, in both cases, we have a
* direct pointer to a task structure thereby bypassing the tasklist.
* We must make sure that, if we have task!= NULL, the target task is still
* present and is identical to the initial task specified
* during pfm_context_create(). It may already be detached from the tasklist but
* that's okay. Note that it is okay if we miss the deadline and the task scans
* the list for nothing, it will affect performance but not correctness.
* The correctness is ensured by using the ctx_lock which prevents the
* notify_task from changing the fields in our context.
* Once holdhing this lock, if we see task!= NULL, then it will stay like
* that until we release the lock. If it is NULL already then we came too late.
*/
LOCK_CTX(ctx);
if (ctx->ctx_notify_task != NULL) {
DBprintk(("[%d], [%d] atomic_sub on [%d] notifiers=%u\n", current->pid,
task->pid,
ctx->ctx_notify_task->pid,
atomic_read(&ctx->ctx_notify_task->thread.pfm_notifiers_check)));
atomic_dec(&ctx->ctx_notify_task->thread.pfm_notifiers_check);
}
if (ctx->ctx_owner != NULL) {
DBprintk(("[%d], [%d] atomic_sub on [%d] owners=%u\n",
current->pid,
task->pid,
ctx->ctx_owner->pid,
atomic_read(&ctx->ctx_owner->thread.pfm_owners_check)));
atomic_dec(&ctx->ctx_owner->thread.pfm_owners_check);
}
UNLOCK_CTX(ctx);
preempt_enable();
pfm_unreserve_session(task, ctx->ctx_fl_system, 1UL << ctx->ctx_cpu);
if (ctx->ctx_fl_system) {
/*
* remove any CPU pinning
*/
set_cpus_allowed(task, ctx->ctx_saved_cpus_allowed);
}
pfm_context_free(ctx);
/*
* clean pfm state in thread structure,
*/
task->thread.pfm_context = NULL;
task->thread.pfm_ovfl_block_reset = 0;
/* pfm_notifiers is cleaned in pfm_cleanup_notifiers() */
}
/*
* function invoked from release_thread when pfm_smpl_buf_list is not NULL
*/
int
pfm_cleanup_smpl_buf(struct task_struct *task)
{
pfm_smpl_buffer_desc_t *tmp, *psb = task->thread.pfm_smpl_buf_list;
if (psb == NULL) {
printk(KERN_DEBUG "perfmon: psb is null in [%d]\n", current->pid);
return -1;
}
/*
* Walk through the list and free the sampling buffer and psb
*/
while (psb) {
DBprintk(("[%d] freeing smpl @%p size %ld\n", current->pid, psb->psb_hdr, psb->psb_size));
pfm_rvfree(psb->psb_hdr, psb->psb_size);
tmp = psb->psb_next;
kfree(psb);
psb = tmp;
}
/* just in case */
task->thread.pfm_smpl_buf_list = NULL;
return 0;
}
/*
* function invoked from release_thread to make sure that the ctx_owner field does not
* point to an unexisting task.
*/
void
pfm_cleanup_owners(struct task_struct *task)
{
struct task_struct *g, *p;
pfm_context_t *ctx;
DBprintk(("called by [%d] for [%d]\n", current->pid, task->pid));
read_lock(&tasklist_lock);
do_each_thread(g, p) {
/*
* It is safe to do the 2-step test here, because thread.ctx
* is cleaned up only in release_thread() and at that point
* the task has been detached from the tasklist which is an
* operation which uses the write_lock() on the tasklist_lock
* so it cannot run concurrently to this loop. So we have the
* guarantee that if we find p and it has a perfmon ctx then
* it is going to stay like this for the entire execution of this
* loop.
*/
ctx = p->thread.pfm_context;
//DBprintk(("[%d] scanning task [%d] ctx=%p\n", task->pid, p->pid, ctx));
if (ctx && ctx->ctx_owner == task) {
DBprintk(("trying for owner [%d] in [%d]\n", task->pid, p->pid));
/*
* the spinlock is required to take care of a race condition
* with the send_sig_info() call. We must make sure that
* either the send_sig_info() completes using a valid task,
* or the notify_task is cleared before the send_sig_info()
* can pick up a stale value. Note that by the time this
* function is executed the 'task' is already detached from the
* tasklist. The problem is that the notifiers have a direct
* pointer to it. It is okay to send a signal to a task in this
* stage, it simply will have no effect. But it is better than sending
* to a completely destroyed task or worse to a new task using the same
* task_struct address.
*/
LOCK_CTX(ctx);
ctx->ctx_owner = NULL;
UNLOCK_CTX(ctx);
DBprintk(("done for notifier [%d] in [%d]\n", task->pid, p->pid));
}
} while_each_thread(g, p);
read_unlock(&tasklist_lock);
atomic_set(&task->thread.pfm_owners_check, 0);
}
/*
* function called from release_thread to make sure that the ctx_notify_task is not pointing
* to an unexisting task
*/
void
pfm_cleanup_notifiers(struct task_struct *task)
{
struct task_struct *g, *p;
pfm_context_t *ctx;
DBprintk(("called by [%d] for [%d]\n", current->pid, task->pid));
read_lock(&tasklist_lock);
do_each_thread(g, p) {
/*
* It is safe to do the 2-step test here, because thread.ctx is cleaned up
* only in release_thread() and at that point the task has been detached
* from the tasklist which is an operation which uses the write_lock() on
* the tasklist_lock so it cannot run concurrently to this loop. So we
* have the guarantee that if we find p and it has a perfmon ctx then it
* is going to stay like this for the entire execution of this loop.
*/
ctx = p->thread.pfm_context;
//DBprintk(("[%d] scanning task [%d] ctx=%p\n", task->pid, p->pid, ctx));
if (ctx && ctx->ctx_notify_task == task) {
DBprintk(("trying for notifier [%d] in [%d]\n", task->pid, p->pid));
/*
* the spinlock is required to take care of a race condition
* with the send_sig_info() call. We must make sure that
* either the send_sig_info() completes using a valid task,
* or the notify_task is cleared before the send_sig_info()
* can pick up a stale value. Note that by the time this
* function is executed the 'task' is already detached from the
* tasklist. The problem is that the notifiers have a direct
* pointer to it. It is okay to send a signal to a task in this
* stage, it simply will have no effect. But it is better than sending
* to a completely destroyed task or worse to a new task using the same
* task_struct address.
*/
LOCK_CTX(ctx);
ctx->ctx_notify_task = NULL;
UNLOCK_CTX(ctx);
DBprintk(("done for notifier [%d] in [%d]\n", task->pid, p->pid));
}
} while_each_thread(g, p);
read_unlock(&tasklist_lock);
atomic_set(&task->thread.pfm_notifiers_check, 0);
}
static struct irqaction perfmon_irqaction = {
.handler = pfm_interrupt_handler,
.flags = SA_INTERRUPT,
.name = "perfmon"
};
int
pfm_install_alternate_syswide_subsystem(pfm_intr_handler_desc_t *hdl)
{
int ret;
/* some sanity checks */
if (hdl == NULL || hdl->handler == NULL) {
return -EINVAL;
}
/* do the easy test first */
if (pfm_alternate_intr_handler) {
return -EBUSY;
}
preempt_disable();
/* reserve our session */
ret = pfm_reserve_session(NULL, 1, cpu_online_map);
if (ret) {
preempt_enable();
return ret;
}
if (pfm_alternate_intr_handler) {
preempt_enable();
printk(KERN_DEBUG "perfmon: install_alternate, intr_handler not NULL "
"after reserve\n");
return -EINVAL;
}
pfm_alternate_intr_handler = hdl;
preempt_enable();
return 0;
}
int
pfm_remove_alternate_syswide_subsystem(pfm_intr_handler_desc_t *hdl)
{
if (hdl == NULL)
return -EINVAL;
/* cannot remove someone else's handler! */
if (pfm_alternate_intr_handler != hdl)
return -EINVAL;
preempt_disable();
pfm_alternate_intr_handler = NULL;
/*
* XXX: assume cpu_online_map has not changed since reservation
*/
pfm_unreserve_session(NULL, 1, cpu_online_map);
preempt_enable();
return 0;
}
/*
* perfmon initialization routine, called from the initcall() table
*/
int __init
pfm_init(void)
{
unsigned int n, n_counters, i;
pmu_conf.disabled = 1;
printk(KERN_INFO "perfmon: version %u.%u IRQ %u\n", PFM_VERSION_MAJ, PFM_VERSION_MIN,
IA64_PERFMON_VECTOR);
/*
* compute the number of implemented PMD/PMC from the
* description tables
*/
n = 0;
for (i=0; PMC_IS_LAST(i) == 0; i++) {
if (PMC_IS_IMPL(i) == 0) continue;
pmu_conf.impl_pmcs[i>>6] |= 1UL << (i&63);
n++;
}
pmu_conf.num_pmcs = n;
n = 0; n_counters = 0;
for (i=0; PMD_IS_LAST(i) == 0; i++) {
if (PMD_IS_IMPL(i) == 0) continue;
pmu_conf.impl_pmds[i>>6] |= 1UL << (i&63);
n++;
if (PMD_IS_COUNTING(i)) n_counters++;
}
pmu_conf.num_pmds = n;
pmu_conf.num_counters = n_counters;
printk(KERN_INFO "perfmon: %u PMCs, %u PMDs, %u counters (%lu bits)\n",
pmu_conf.num_pmcs,
pmu_conf.num_pmds,
pmu_conf.num_counters,
ffz(pmu_conf.ovfl_val));
/* sanity check */
if (pmu_conf.num_pmds >= IA64_NUM_PMD_REGS || pmu_conf.num_pmcs >= IA64_NUM_PMC_REGS) {
printk(KERN_ERR "perfmon: not enough pmc/pmd, perfmon disabled\n");
return -1;
}
/*
* for now here for debug purposes
*/
perfmon_dir = create_proc_read_entry ("perfmon", 0, 0, perfmon_read_entry, NULL);
if (perfmon_dir == NULL) {
printk(KERN_ERR "perfmon: cannot create /proc entry, perfmon disabled\n");
return -1;
}
/*
* create /proc/perfmon
*/
pfm_sysctl_header = register_sysctl_table(pfm_sysctl_root, 0);
/*
* initialize all our spinlocks
*/
spin_lock_init(&pfm_sessions.pfs_lock);
/* we are all set */
pmu_conf.disabled = 0;
return 0;
}
__initcall(pfm_init);
void
pfm_init_percpu(void)
{
int i;
int me = get_cpu();
if (me == 0)
register_percpu_irq(IA64_PERFMON_VECTOR, &perfmon_irqaction);
ia64_set_pmv(IA64_PERFMON_VECTOR);
ia64_srlz_d();
/*
* we first initialize the PMU to a stable state.
* the values may have been changed from their power-up
* values by software executed before the kernel took over.
*
* At this point, pmu_conf has not yet been initialized
*
* On McKinley, this code is ineffective until PMC4 is initialized.
*/
for (i=1; PMC_IS_LAST(i) == 0; i++) {
if (PMC_IS_IMPL(i) == 0) continue;
ia64_set_pmc(i, PMC_DFL_VAL(i));
}
for (i=0; PMD_IS_LAST(i); i++) {
if (PMD_IS_IMPL(i) == 0) continue;
ia64_set_pmd(i, 0UL);
}
put_cpu();
pfm_freeze_pmu();
}
#else /* !CONFIG_PERFMON */
asmlinkage long
sys_perfmonctl (int pid, int cmd, void *req, int count, long arg5, long arg6,
long arg7, long arg8, long stack)
{
return -ENOSYS;
}
#endif /* !CONFIG_PERFMON */
|