I cranked out a little test application this weekend to look at problems in the RT scheduler. You can find the source here. ftp://ftp.novell.com/dev/ghaskins/preempt.cc Here is some example output: # ./a.out -d 40 -a Starting test with 5 threads on 4 cpus Calibration: 40ms = 11306781 loops 0 -> p:2 a: 1 l: 5/4 t: 32788/32847/59 1 -> p:5 a: 0 l: 8/9 t: 34401/34532/131 2 -> p:4 a: 3 l: 5/29 t: 34659/34745/86 3 -> p:3 a: 2 l: 8/7 t: 34852/34925/73 4 -> p:1 a: 0 l: 42/9 t: 74073/74116/43 --------------------------- Stats: [min: 4us @ row 0] [max: 42us @ row 4] {ave: 12us] --------------------------- Test PASSED ------------------------------ Here is the meaning for each row, by example of row 0: 0 -> p:2 a: 1 l: 5/4 t: 32788/32847/59 This says the first thread to exit was priority-2, affined to cpu 1, with a latency of 5us/4us (child-start, parent-return), for a local-relative run-time of 32.788ms, a global-relative runtime of 32.847ms, with a delta between the two of 59us. The list is sorted by finish time. So you can see that the priority 2 thread finished first, priority 5 finished second, 4 third, etc. The test considers a "PASS" to be that the top N threads (where N is nr_cpus) finish first, the second group finish second, etc. For instance, on a 4-core system with 10 threads, 10,9,8,7 should finish first (though their ordering within that group may vary). 6,5,4,3 should finish second. And finally 2,1 should finish last. Any violation of this ordering will be flagged as follows: # ./a.out -t 95 -a Starting test with 95 threads on 4 cpus Calibration: 20ms = 5686503 loops 0 -> p:94 a: 1 l: 6/93 t: 14809/20098/5289 1 -> p:92 a: 3 l: 6/86 t: 15217/20367/5150 2 -> p:95 a: 2 l: 7/83 t: 15064/20455/5391 3 -> p:93 a: 0 l: 6/36 t: 19452/24696/5244 ..<snip>... 79 -> p:15 a: 2 l: 5/57 t: 376806/378567/1761 80 -> p:16 a: 3 l: 6/59 t: 376838/378665/1827 81 -> p:13 a: 0 l: 6/17 t: 385565/387280/1715 82 -> p:14 a: 1 l: 6/13 t: 389003/390743/1740 83 -> p:11 a: 2 l: 9/14 t: 396074/397693/1619 84 -> p:12 a: 3 l: 5/67 t: 397287/398927/1640 85 -> p:9 a: 0 l: 6/6 t: 405105/406684/1579 86 -> p:10 a: 1 l: 8/12 t: 405816/407411/1595 87 -> p:7 a: 2 l: 8/14 t: 416301/417786/1485 88 -> p:8 a: 3 l: 4/66 t: 417852/419357/1505 89 -> p:6 a: 1 l: 6/10 t: 423228/424694/1466 90 -> p:5 a: 0 l: 8/9 t: 424050/425500/1450 91 -> p:3 a: 2 l: 4/8 t: 434555/435938/1383 92 -> p:4 a: 3 l: 6/33 t: 435763/437163/1400 93 -> p:2 a: 1 l: 8/8 t: 442051/443420/1369 94 -> p:1 a: 0 l: 1349/8 t: 443747/445098/1351 --------------------------- Stats: [min: 4us @ row 88] [max: 1349us @ row 94] {ave: 27us] --------------------------- Error: row 79: 15 != (19 >= x > 15) Error: row 80: 16 != (15 >= x > 11) Error: row 83: 11 != (15 >= x > 11) Error: row 84: 12 != (11 >= x > 7) Error: row 87: 7 != (11 >= x > 7) Error: row 88: 8 != (7 >= x > 3) Error: row 91: 3 != (7 >= x > 3) Error: row 92: 4 != (3 >= x > 0) --------------------------- Test FAILED with 8 errors --------------------------- This output shows, for instance, that row 79 has a priority 15 thread that finished when only 19,18,17, or 16 should have if the system was prioritizing properly. Comments/suggestions welcome! Regards, -Greg - To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
