Hi Minos,
Is this a compatibility issue and the TFM doc build system can be upgraded, or this a bug in the latest version of Sphinx?

Hi Ken,
Yes, the final patch has fixed the issue.
It works now without our workarounds.

Should to confirm, that after the patch the issue is still present.

Hi Ken,
after farther testings, PSA Test application still goes to the memory fault.
So I have returned all my two workarounds, which work for all applications:

1. .stack_top = PART_REGION_ADDR(ARM_LIB_STACK, $$ZI$$Limit) - 0x48,
2. Disable if (stacked_ctx_pos != p_cur_sp->runtime_data.sp_thrd.sp_btm)

Hi Ken,
Also, it should be done something with the check in tfm_svcalls.c, tfm_core_validate_caller() line 1001:

Hi Ken,

So my workaround, to avoid psa_panic(), is to disable the check:

So I have returned to the RESET issue.

After the following fix - 0x48, the applications does not have the memory fault:

{
    .stack_bottom         = PART_REGION_ADDR(ARM_LIB_STACK, $$ZI$$Base),
    .stack_top            = PART_REGION_ADDR(ARM_LIB_STACK, $$ZI$$Limit) - 0x48, //NXP
    .rw_start             = PART_REGION_ADDR(ARM_LIB_STACK, $$ZI$$Base),
},

Hi Ken,

I have tried to workaround it. Current steps:

1. Updated to the latest TFM commit.
2. Reverted changes related to:
   1. 5248af2d7b86775364a0e131eb80ac0330bc81fb
   2. 490281df3736b11b62e25bc98d3e2c6e4e10478c
   3. 483da6447e4360c514538807275904be68395dff

After that, PSA Test application starts working.
But the regression test application still goes to the Memory fault during the first App ROT partition initialization, on the first SVC call.

3. Switched App ROT partitions to PSA ROT, in tfm_spm_db.inc => after that regression tests are OK.

Hi Antonio
FYI:
mbedCrypto v3.0.1: https://github.com/ARMmbed/mbed-crypto/tags

FYI:
For ARMClang/Keil

1. Functional API and TFM_LVL=1, all applications/tests are OK!!!
2. IPC API and TFM_LVL=1 => repetitive reset.
3. IPC API and TFM_LVL=2 => goes to the mentioned memory fault.

Another experiment. Switching to Armasm V5 from v6 did not help, the same result - goes to the Memory fault.
So, for this moment only GCC (for PSA tests) is OK.

BTW: At the same time the TFM Regression Tests application (using GCC) goes to infinite reset on test_app()=>tfm_secure_client_run_tests()=>psa_connect(TFM_SECURE_CLIENT_SFN_RUN_TESTS_SID,TFM_SECURE_CLIENT_SFN_RUN_TESTS_VERSION)=>tfm_ns_interface_dispatch()=>fn()=>causes reset

Hi Ken,

Hi Ken,

After tfm_core_svc_handler() returns, the execution jumps to tfm_nspm_thread_entry().
Strange is, that during this jump PSP is changed from 0x30000FE0 to 0x30001048.

So the PSP, during main() has following values:
PSP = 0x300001000 PSPLIM = 0x30000800

Hi Ken,
Thank you by taking care about it.
Guess, the issue may be caused by the way we are using TFM. We have to compile TFM Secure in one project (instead of separate libraries). Guess, it may cause that some parts of data/code may be placed to unwanted memory sections.
I will try to analyze the generated map file.

Hi Antonio, it looks like the patch is forgotten by reviewers.

The separation of trees does not help without changes in original code.
#include "psa/<same_name.h>" will reference to different versions of file (in case two library sources are in one project) .
The only way to solve it, is to change #include "differen_name_or_path" (our solution), or compile and use MbedCrypto as a separate library (your solution).
Adding a file prefix or changing of the "psa" sub-folder name to something different, will solve the file name conflict.

It looks like there is no conflict, but it is ;)
TFM and MbedCrypto referencing to "psa/<same_name.h>" which are placed in different directories.
But if both sources belong to the same project, the project must contain both include paths .
It is a luck what path is taken the first one in the search path list.
If no luck, you will get compilation errors as some structure types have an unexpected content.

Symbol renaming is OK.
But I have talked about file names, to avoid conflict when TFM and MbedCrypto referencing to "psa/<same_name.h>".
Guess, using a different include directory e.g. "psa_tfm/<same_name.h>" may solve the issue.
P.S. Different headers files should have different names and path - good practice for any use case.

It means that you have decided to add the limitation which is possible to avoid.
It is possible to have and compile TFM, MbedCrypto and PSA test suite without splitting to separate libraries/projects.

It's probably OK when to compile in multiple libraries. The problem happens when to compile sources of MbedCrypto and TFM in one project.
So, as workaround, we have to add psa_ prefix to TFM PSA headers (as it was done in previous version of TFM).

Hi, PSA header files in mbedCrypto and TFM have the same names but a different content.
Guess, if they have the same names and the same path, they should to have the same content or to have a different prefix.

Thanks, hope it will be implemented.