Did you try with #71593 ?

Did you try with #71593 ?

Yes

I realized that it fails to boot on warm reboots (cold reboots are fine).

It appears that I have missed something important around the (seemingly inconsistent) use of BASEPRI & cpsie/d… 🤔

Interrestingly, this isn’t caught by any of the tests I ran. I would like to fix the issue but also eventually add some tests to capture future regressions.
Could you please detail the process you’re following to reproduce the issue?
Do you use an in-tree example/test? How do you load/reload/reset?

I also encountered a similar phenomenon with arduino_uno_r4_minima (Renesas RA4M1, Cortex-M4/ARMv7-M).
Similar to this issue, I confirmed that reverting f11027d works correctly.
It is also working fine with @duda-patryk patch.

Thank you for letting me know. I am currently running tests on a nucleo_f401re but can’t seem to be able to reproduce the issue.

Along with the fix, I’d like to understand & document the why is this extra cpsie required.

Have you tried jumping to reset vector with interrupts disabled?

I am running Zephyr’s tests & samples.
I get some false positives (ie failing test because of twister missing the start of the log) but they match before & after the changes to arch/arm.

Would you be able to provide a sample project that exhibits this issue?

In my case, I'm jumping to Zephyr reset vector from another code (chain-loading) with interrupts disabled (cpsid i).

To reproduce issue without chain-loading you can use following code:

void z_arm_reset(void);

void test(void)
{
        __disable_irq();
        z_arm_reset();
}

So I edited samples/hello_world/src/main.c such that it becomes:

#include "zephyr/kernel.h"
#include <stdio.h>

extern void z_arm_reset(void);

int main(void) {
    printf("Hello World! %s\n", CONFIG_BOARD_TARGET);

    k_sleep(Z_TIMEOUT_MS(2000));

    __disable_irq();
    z_arm_reset();

    /* should not be reached */
    printf("Reboot failed\n");
    k_sleep(Z_FOREVER);

    return 0;
}

And this indeeds behaves incorrectly as the sleep is not stopping the thread for the requested amount of time.

However, z_arm_reset is not a proper way to reset the system, sys_reboot(_) from zephyr/sys/reboot.h is. Calling sys_reboot(SYS_REBOOT_WARM) or sys_reboot(SYS_REBOOT_COLD) does work as expected.

With reguards to chain loading, I also tried MCUBoot with the helloworld (using sys_reboot(SYS_REBOOT_WARM) and sys_reboot(SYS_REBOOT_COLD)) and it all seems to work as expected too.

For the record, here’s how I built & flashed them:

west build --sysbuild -p -b nucleo_f401re samples/hello_world -DSB_BOOTLOADER=MCUBOOT=y
pyocd load build/mcuboot/zephyr/zephyr.elf
pyocd load build/hello_world/zephyr/zephyr.signed.hex

It might be worth noting that the documentation for BOOTLOADER_MCUBOOT states that:

 By default, this option instructs Zephyr to initialize the core architecture HW registers during boot, when this is supported by the application. This removes the need by MCUboot to reset the core registers' state itself.

Could this be something useful for your usecase?

MCUBoot works because they use irq_lock() to disable interrupts (faults are enabled) https://github.com/mcu-tools/mcuboot/blob/d2e69bf720ec4e5825f1bd7be441a9e3dedb5533/boot/zephyr/main.c#L205

@yperess do you remember why we disable interrupts using cpsid i?

CONFIG_MCUBOOT_CLEANUP_ARM_CORE defaults to y.

When the cleanup is enabled, do_boot() calls cleanup_arm_nvic() which disables interrupts too 🤔.

When the cleanup is enabled, do_boot() calls cleanup_arm_nvic() which disables interrupts too

__disable_irq() is called indeed. @ithinuel I assume that it works for you 😅

Well, I cannot reproduce the issue but from the bug report what's missing is enabling the IRQ so if anything it only confuses me more.

Are you happy closing this issue with the fix that was merged?

Closing