MCS: Add periodic drift test #76
Conversation
If this test fails, TIMER_OVERHEAD_TICKS may not be set correctly for the platform. Needs seL4 pull #847 applied to pass. Signed-off-by: Indan Zupancic <[email protected]>
|
Without seL4/seL4#847 this fails with 2498 and 2494 for SMP instead of the expected 2500 on tqma8xqp1gb. |
There was a problem hiding this comment.
If I understand this right, the main thread sleeps for (1 second - half-a-period = 1 - 0.000002s) which should guarantee that each of the child threads should be allowed to yield exactly (1 second / period = 2500) times. This means that if a period drifts by more than 80ns for each activation on average the test will fail, but if it's less than 80ns it will pass?
|
Correct. The test can be made less strict by reducing the sleep time. |
| /* Create tasks with lower priority: */ | ||
| for (int i = 0; i < num_threads; i++) { | ||
| create_helper_thread(env, &t[i]); | ||
| set_helper_priority(env, &t[i], env->priority - 1); |
There was a problem hiding this comment.
maybe add an assert here that this does not cause a roll over.
There was a problem hiding this comment.
Good idea. test_timeout_fault_in_server does OUR_PRIO - 2, and #define OUR_PRIO (env->priority). It seems priority is set to seL4_MaxPrio - 1 in main.c.
@Indanz Could you add this as a comment that describes the test's details? |
If this test fails, TIMER_OVERHEAD_TICKS may not be set correctly
for the platform. Needs seL4 pull #847 applied to pass.
Signed-off-by: Indan Zupancic [email protected]