Hi,
We have seen random ARM faults.
After some time we have seen that the lr register often pointed to entropy_cc310_rng_get_entropy.
This is our setup:
- Environment: Zephyr, SDK 1.2
- An IRQ giving a semaphore to a thread A
- A thread A with higher priority than thread B waiting for a semaphore given by the IRQ.
- A thread B with a lower priority than thread A using a function that boils down to a spm_xxx call such as spm_request_random_number.
The problem arises when we get an interrupt and zephyr switches to the higher priority thread (thread A) while thread B executes the spm_xxx function.
I have enclosed a simple project that in general fails within a couple of hundred counter interrupts. The project runs fine if the following modification is done to the entropy_cc310 driver in the nrf repo:
diff --git a/drivers/entropy/entropy_cc310.c b/drivers/entropy/entropy_cc310.c
index 6888b218..e471e428 100644
--- a/drivers/entropy/entropy_cc310.c
+++ b/drivers/entropy/entropy_cc310.c
@@ -34,7 +34,9 @@ static int entropy_cc310_rng_get_entropy(struct device *dev, u8_t *buffer,
/** This is a call from a non-secure app that enables secure services,
* in which case entropy is gathered by calling through SPM
*/
+ k_sched_lock();
res = spm_request_random_number(buffer, length, &olen);
+ k_sched_unlock();
if (olen != length) {
return -EINVAL;
}
Thank you for looking into this.
best regards,
Daniel.
