Repository navigation
ThreadX modules support on Cortex-M33 #431
Description
I am running ThreadX from tag v6.4.1_rel with modules-enabled kernel.
When compiled with TX_SINGLE_MODE_SECURE - it hard faults after first enter to the scheduler and PendSV interrupt.
PendSV interrupt executes until returning from exception.
EXC_RETURN was set to 0xFFFFFFBC and that was causing aforementioned hardfault.
After applying patch below - scheduler runs successfully.
diff --git a/ports_module/cortex_m33/gnu/module_manager/src/tx_thread_schedule.S b/ports_module/cortex_m33/gnu/module_manager/src/tx_thread_schedule.S
index e392380b..ba8c5fea 100644
--- a/ports_module/cortex_m33/gnu/module_manager/src/tx_thread_schedule.S
+++ b/ports_module/cortex_m33/gnu/module_manager/src/tx_thread_schedule.S
@@ -471,6 +471,12 @@ _skip_vfp_restore:
LDMIA r12!, {r4-r11} // Recover thread's registers
MSR PSP, r12 // Setup the thread's stack pointer
+#ifdef TX_SINGLE_MODE_SECURE
+ LDR r0, =0xFFFFFFFD // Exception return to secure
+#else
+ LDR r0, =0xFFFFFFBC // Exception return to non-secure
+#endif
+ MOV lr, r0
BX lr // Return to thread!
It might be that I am not understanding Trustzone secure/non-secure implications.
Following addition was present in MemManage handler e.g. - other interrupts do not have it.
Could someone suggest if this is a scheduler bug for CM33/CM23 and other CMSE architectures or I am configuring my microcontroller wrongly?
Activity
fdesbiens commentedon Jan 24, 2025
Thank you for this question, @Jasinsky.
@eclipse-threadx/iot-threadx-committers and @eclipse-threadx/iot-threadx-contributors: can one of you please comment on this issue?
fdesbiens commentedon Sep 23, 2026
I confirmed that this was a bug in
v6.4.1_rel. The GNU Cortex-M33 module manager’stx_thread_stack_build.scontained theTX_SINGLE_MODE_SECUREconditional, but GAS did not preprocess the lowercase.sfile. Both branches were therefore assembled, and0xFFFFFFBCoverwrote the secure0xFFFFFFFDvalue in the initial thread frame.I fixed this on
devin #672 by renaming the affected assembly source to.S, and I added checks that assemble the GNU module ports with their feature macros enabled. I do not recommend overridinglrin the scheduler because it must retain the savedEXC_RETURNproperties, including the floating-point frame state.I am closing this as fixed. Thank you for reporting it and for providing the diagnosis.