diff options
Diffstat (limited to 'test/smp/regression/threadx_thread_priority_change.c')
| -rw-r--r-- | test/smp/regression/threadx_thread_priority_change.c | 50 |
1 files changed, 46 insertions, 4 deletions
diff --git a/test/smp/regression/threadx_thread_priority_change.c b/test/smp/regression/threadx_thread_priority_change.c index e68eadf5..d22f3d72 100644 --- a/test/smp/regression/threadx_thread_priority_change.c +++ b/test/smp/regression/threadx_thread_priority_change.c @@ -465,11 +465,28 @@ TX_THREAD *temp_thread; /* Restore interrupts. */ TX_RESTORE - /* Hit branch in _tx_thread_smp_simple_priority_change such that the - new priority is less than the inheritance priority. */ - thread_0.tx_thread_inherit_priority = 0; + /* Hit the branch in _tx_thread_smp_simple_priority_change where the new + priority is below the inheritance priority. That needs an inheritance + priority above the new one, so 0 does not reach it: with 0 the other + branch runs, and that branch records the inheritance priority as the + thread's priority while still linking the thread at the new priority. + Thread 0 came out of here claiming priority 0 while living in the + priority 16 list, which is what has been hanging this test in CI. + Priority 0 ties with the control thread, so resuming the control thread + raised no preemption, and once thread 0 completed, core 0 was left owned + by a completed thread with the control thread ready and unscheduled. + + This routine is internal and expects protection to be in force, so hold + it here rather than calling in the open. */ + TX_DISABLE + thread_0.tx_thread_inherit_priority = 20; _tx_thread_smp_simple_priority_change(&thread_0, 16); + /* Put the inheritance priority back to the value that means this thread is + not inheriting one, so nothing downstream reasons about a phantom. */ + thread_0.tx_thread_inherit_priority = TX_MAX_PRIORITIES; + TX_RESTORE + /* Successful test. */ printf("SUCCESS!\n"); test_control_return(0); @@ -513,6 +530,15 @@ static void thread_3_entry(ULONG thread_input) UINT old_priority; UINT loop; +/* The interrupt below only fires on a narrow window: thread 3 at priority 6, + ready, and not yet at the head of its priority list. Waiting for it without a + bound made this test hang indefinitely when the window was never hit, which + showed up in CI as a timeout carrying no information about what went wrong. + The TX_NOT_INTERRUPTABLE path already stopped after a fixed amount of work; + this gives the interruptable path the same protection. */ +UINT attempts = 0; +#define MAX_PRIORITY_CHANGE_ATTEMPTS 100000 + /* Resume threads 4. */ tx_thread_resume(&thread_4); @@ -537,7 +563,23 @@ UINT loop; if (thread_4_counter) thread_1_counter++; - } while (test_isr_dispatch); + attempts++; + + } while (test_isr_dispatch && (attempts < MAX_PRIORITY_CHANGE_ATTEMPTS)); + + /* Stop the interrupt regardless, so a window that was never hit does not + leave the handler installed for the rest of the test. */ + test_isr_dispatch = TX_NULL; + + if (attempts >= MAX_PRIORITY_CHANGE_ATTEMPTS) + { + + /* The window was not reached. That is a gap in what this test covered on + this run rather than a fault in the code under test, so say so instead + of failing, and leave the counters alone so the checks that follow + still mean what they did before. */ + printf("(priority change window not reached in %u attempts) ", attempts); + } } |