Failed tests: posix.sema and posix_checks on em_starterkit_em7d_v22

[nniranjhana]

Test in tests/posix/semaphore failing on emsk platform with Stack Check Fail error:

***** delaying boot 1000ms (per build configuration) *****
***** Booting Zephyr OS 1.12.99 (delayed boot 1000ms) *****
Running test suite test_sem
===================================================================
starting test - test_sema
Exception vector: 0x6, cause code: 0x2, parameter 0x2
Address 0x80002800
***** Stack Check Fail! *****
Current thread ID = 0x800068a8
Fatal fault in thread 0x800068a8! Aborting.

Test in tests/posix/posix_checks failing with similar error after starting:

***** delaying boot 1000ms (per build configuration) *****
***** Booting Zephyr OS 1.12.99 (delayed boot 1000ms) *****
Running test suite test_posix_checks
===================================================================
starting test - test_sched_policy

The given policy number 0 is supported. The minimum priority is 0 and the maximum priority is 15.
The given policy number 1 is supported. The minimum priority is 0 and the maximum priority is 15.
Exception vector: 0x6, cause code: 0x2, parameter 0x2
Address 0x80001800
***** Stack Check Fail! *****
Current thread ID = 0x800058a8
Fatal fault in thread 0x800058a8! Aborting.

[vonhust]

@ruuddw @andrewboie @nashif

The bug is introduced in kernel: fix errno access for user mode (commit 7f4d006).

For ARC' stack check, Non-%SP based memory references may not reference the unallocated stack region, so the access to errno will raise stack check exception.

[vonhust]

@ruuddw @andrewboie @nashif
For
new_thread->errno_location = (int *)K_THREAD_STACK_BUFFER(stack);

I think it's better to use the following

new_thread->errno_location = (int *)(new_thread->stack_info.start + new_thread->stack_info.size);

Because K_THREAD_STACK_BUFFER(stack) may be not the top of thread stack.
In ARC, for kernel thread, the top of kernel thread stack is K_THREAD_STACK_BUFFER(stack) - MPU_STACK_GUARD_SIZE.

For arch, i think it's possible to make adjustments on start address and size, based on the defined stack and size to meet the requirement of mpu address alignment, e.g. 33 bytes of stack aligned to 64 bytes of stack .

So the information in thread->stack_info is the final, correct, adjusted information about thread stack.

[andrewboie]

I can take this, I'll move the errno location to the other end of the stack.

[ruuddw]

#9264 fixes it by excluding the errno from the stack checking bounds (though some review comments stil pending on that one). Moving errno to the other end probably solves it as well.

[andrewboie]

I think we should move it to the other end of the stack, it's a safer solution for all arches, the current design could introduce corruption if the stack pointer gets pushed up to, but does not exceed, the top.