ta: os_test: fetch system time after the one second delay for proper comparaison #703
Conversation
ta/os_test/os_test.c
Outdated
| TEE_GetSystemTime(&sys_t); | ||
| printf("system time %u.%03u\n", (unsigned int)sys_t.seconds, | ||
| (unsigned int)sys_t.millis); | ||
|
|
||
| if (t.seconds > sys_t.seconds) { |
There was a problem hiding this comment.
This test seems a bit strange. What are we testing against here, a time that has wrapped too much?
There was a problem hiding this comment.
I guess this test checks if the TA time wrapped correctly?
If the TA time wraps correctly after the one second delay, the TA time should be inferior to one second. That TA time will always be inferior to the system time (with that patch).
If the TA time does not wrap and stays to its max value, the TA time will be superior to system time. There is still a chance where TA time and system time are at the maximum value at the very same time ... but that's unlikely.
There was a problem hiding this comment.
I guess the test checks that TA time has indeed wrapped. Maybe testing against 0 should be sufficient.
It seems the issue is rather that System Time got reset during a low power state transition whereas it should not. |
I agree that a system counter reset should not be expected, we are investigating. However, I still think the logic of this test is wrong. Let's imagine you system boots super-fast and run xtest, this test will fail. |
|
I agree that there is no need to test TA wrapped time against TEE system time. We just need to test it wrapped. As we expect only few seconds have passed, testing t_seconds < 2 should be enough, or maybe 5. |
clementfaure
force-pushed
the
upstream/fix_sys_time
branch
from
a3b45fe to
23089c0
Compare
clementfaure
force-pushed
the
upstream/fix_sys_time
branch
from
23089c0 to
6c4217b
Compare
|
Tag applied |
|
|
In the TA time wrap test, the TA time is set to its maximum value. After one second delay, the TA time was compared to the system time. The comparaison with the system time might not be pertinent as it is most likely more than dozen seconds. Compare the wrapped time to a fixed value instead. Signed-off-by: Clement Faure <[email protected]> Acked-by: Etienne Carriere <[email protected]> Acked-by: Jens Wiklander <[email protected]>
clementfaure
force-pushed
the
upstream/fix_sys_time
branch
from
6c4217b to
89cc1d7
Compare
|
Tags applied |
…time comparaison
In the regression_1006 test, the wrapped TA time is compared to the system time. The time delay for this test is one second. The system time is fetched at the very begining of the test function.
For this test to pass, it is assumed that the system time fetched is at least one second. That might not be the case in a system where the test is started right after a deep sleep state where the system counter gets reset.
To solve this issue, fetch the system counter after the one second delay.